What

Bumps the pinned markup_fmt dependency from =0.27.0 to =0.27.2
(Cargo.toml and Cargo.lock), plus a one-line type fix in cli/tools/fmt.rs
that the new version requires.

Why

deno fmt did not terminate on large HTML files with many inline style="..."
attributes (#30982; a 10 MB file ran for 15+ minutes with no output).

The cause is in markup_fmt 0.27.0. FormatCtx::format_style_attr
(src/ctx.rs) prepended a whitespace-masked copy of the entire preceding
document (self.source[0..start]) to the few bytes of inline CSS before handing
it to the external CSS formatter (malva). Since start grows linearly down the
file, each of the N style attributes does O(start) work, which is O(n^2) overall.
That prefix only existed to keep byte offsets aligned for error reporting, and
the style-attr result is trimmed and uses print_width = u16::MAX, so the prefix
never affects the output.

markup_fmt 0.27.2 (g-plane/markup_fmt#223, merged upstream) drops the prefix
and instead computes source line/column lazily, only on error, via
pos_to_line_col. That makes formatting linear again. The fix is already
released upstream, so this PR advances Deno's pin to pick it up.

How

• Cargo.toml: change markup_fmt = "=0.27.0" to "=0.27.2".
• Cargo.lock: regenerated with cargo update -p markup_fmt --precise 0.27.2.
  0.27.2 also pulls in memchr 2.7.5 to 2.8.1 and adds tiny_pretty 0.4.2;
  both are markup_fmt's own declared deps.
• cli/tools/fmt.rs: change the format_embedded_html closure error type from
  std::convert::Infallible to deno_core::anyhow::Error (see below).

Source change required by 0.27.2

0.27.2 changed format_text from Result<_, FormatError<E>> to a concrete
Result<_, FormatError>, where the external formatter closure must return an
error type that converts into anyhow::Error. Deno has two call sites:

• format_html (line 451) already returns AnyError
  (deno_core::error::AnyError = anyhow::Error), so it needed no change.
• format_embedded_html (line 750) returned Ok::<_, std::convert::Infallible>,
  which no longer satisfies the bound. I changed it to
  Ok::<_, deno_core::anyhow::Error> to match the other site.

My first pass only checked the format_html closure and missed this second one,
so the initial commit broke the debug build. CI caught it at fmt.rs:751; this
is now fixed and cargo check -p deno compiles clean.

Verification / benchmark

I reproduced this end to end against Deno's exact pinned crate versions (stock
0.27.0 vs 0.27.2 via [patch.crates-io]), formatting HTML with N
<div style="..."> lines:

Stock is quadratic; the bumped version is linear (about 113x faster at N=8000),
and the formatted output is byte-identical.

Checklist (PR template)

• Descriptive conventional-commit title (fix(fmt): ...).
• Related issue referenced (Closes #30982).
• ./x fmt: cargo fmt --check passes on cli/tools/fmt.rs (edition 2024),
  and dprint check Cargo.toml passes.
• ./x lint: cargo check -p deno compiles clean after the closure fix.
• Tests: added tests/specs/fmt/html_many_inline_styles, a fmt spec that
  runs fmt --check on an HTML file with 150 inline style="..." divs and
  asserts it completes. It passes in well under a second against a local
  build; on 0.27.0 the same input does not terminate.

AI usage disclosure

This PR was prepared with help from an AI coding agent. I reviewed and verified
the diagnosis, the upstream root cause (g-plane/markup_fmt#223), the version
bump, the closure type fix, and the benchmark locally.