feat(minify): close more oxc and swc gaps, keep with bodies free of flattened names - #22919

…te them

Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP

…und (#22903)

* fix(html): cover the printer from its own tests and fix what that found

Prints that re-parsed to a different tree, now each printed so the tree
comes back the same:
- content after a `<plaintext>` no longer gets end tags that read as
  text; a script left double-escaped at EOF gets no `</script>`;
- a `<?pi>` keeps the shell tags around it, as a comment does;
- `<body>` stays before a head element, a `<noscript>`, a template or
  whitespace collapsing removes; `<html>` in foreign content keeps its
  start tag; `<tbody>`/`<colgroup>` in template content and `</caption>`
  before a script are kept;
- nested forms, text fostered out of a table and merged into earlier
  text, and fragment prints (whitespace under the root, a stray `<html>`,
  non-body contexts) are re-parsed and fall back to the source.
A streamed walk starting after a `</form>` no longer visits out of
order, `skipChildren()` while printing no longer writes `undefined`, and
`<tr> <table><tr><option><li>` no longer throws.

The printer tests now reach every statement and branch of the HTML
printer and minifier; checks no input could fail are folded away.

Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP

* fix(html): cover the insertion modes from the parser tests too

The parser tests' earlier reach left out the insertion modes (`modes.*`),
which are assigned after `const modes = {}`; they now reach every
statement and branch of those too, and two trees were wrong:
- an `<a>` start tag ran the adoption agency, then removed the first
  `a` anywhere in the active formatting list, which can be one below a
  marker: in `<a>1<marquee><a>2<a>3</marquee>4` the outer link was lost.
- in a head `<noscript>` with scripting off, the whitespace before other
  text went to the head after it; the mode now splits it off first, as
  in head and after head do.
Checks the stack and list invariants rule out (an `<html>` at the bottom
of every scope, a formatting element still listed after the inner loop,
a detached node always among its parent's children, …) are folded away.

Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP

* fix(css): cover the printer from its own tests and fix what that found

Prints that changed what a sheet means, or threw:
- `background-position` with an empty layer printed `undefined` and
  turned a value the browser drops into one it applies;
- a `url()` holding two strings was rebuilt as one valid url;
- `@supports ((display:grid) {…}` cut short lost its outer `(`, so the
  block opened and the rule applied;
- box longhands holding a word the longhand does not read were merged
  into a shorthand the browser drops whole (and the shorthand lowered
  into partly valid longhands);
- `escapeIdentifier` wrote U+0001–U+001F/U+007F raw and a lone `-` bare;
  `normalizeUrl` decoded `\1F600` as U+F600 and `\110000` as U+0000;
- nesting lowered inside a block big enough to stream lost the hoisted
  rules (an empty sheet for a 4000-rule block);
- `skipChildren()` while printing, `color-mix()` with an empty argument
  and `as: "block-contents"` with a nested or streamed block threw.

The printer tests now reach every statement and branch of
syntax-printer.js and every printer function in syntax-parser.js; checks
no input could fail are folded away.

Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP

* fix(css): cover the rest of the printer and fix what that found

Ordered and deduplicated layer gathers, empty @layer blocks no longer
stop later joins, value-less grid longhands no longer merge, and an
unquoted url no longer reads as a comment when scanning layer text.

Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP

* fix(javascript): cover the printer from its own tests and fix what that found

Nested IIFE run-in-place crash, uncompilable RegExp() under unsafe,
folding patched builtins, global_defs on an exported binding,
RegExp.$1 under unsafe, ESTree shorthand __proto__, and spidermonkey
directive values.

Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP

* fix(javascript): reach more of the printer and fix what that found

A join of one hole, hoist_vars across a block with let, a computed
__proto__ key under unsafe, and new RegExp(/a/g) losing its flags.

Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP

* chore: add the printer changeset and fix the type check

Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP

* fix(css): lower an inset of a lone CSS-wide keyword again

The box-slot check refused `inset:inherit`, which every longhand reads;
a keyword beside another value still stays, as the target drops it.

Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP

* fix(javascript): keep the literal a built-in method is called on

A computed read like `[1]['concat']` rewritten to a dot read lost its
receiver's elements, as the call guard compared the rewritten read with
the one its parent calls.

Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP

* test(javascript): read a duplicate-named regular expression by the runtime

Duplicate group names in alternatives compile on newer runtimes only, so
whether its `source` folds follows the runtime the test runs on.

Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP

* fix(css): tell a quoted url from an unquoted one past leading whitespace

The lookahead backtracked into the whitespace, so `url( "a)b")` read
as unquoted.

Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP

* fix(javascript): read the global object as handed on only where it may be, and run the printer tests on Bun

A binding nothing reads, a literal's method and a module's function leak
no global object, which kept swc's corpus bigger. The printer tests
compare runtime output only where the engine compiles the source, read
engine errors by type, and skip what a small stack stops.

Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP

* perf(html): read back only the fragments a body does not read alike

Every fragment's print was parsed again to check it, which doubled the
cost of one in a `tr`; foreign, `select` and raw-text contexts still are.

Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP

* test(javascript): leave acorn's deepest program out where Bun's stack is small

Bun runs a test in a worker whose stack no parser gets 2000 levels deep
in, which every corpus test now reads from one list.

Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP

* fix(javascript): keep a nested index map's section that ends past the outer one

A section's stop clamped the outer stop's column to the next section's
even where that one started on a later line, which cut a nested map's
first source to nothing; terser's source-map reader does the same.

Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP

* test(css): read a visitor's replace and remove through the printer's own tests

#22906's edits are written out by the printer, which its own tests now
read, a replaced node holding a replaced one included.

Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP

* test(javascript): take the printer's snapshots as main's new passes write them

Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP

* fix(css): read past the comments after a visitor stops the walk

stop() drops the comment bucket while the node it is in still tokenizes,
so on a walk that prints nothing the next comment read a bucket it no
longer had and threw. The guard folded as dead is back, with its tests.

Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP

* test(javascript): take the printer's snapshots as #22919's passes write them

Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP