* feat: introduce format.zig with InputFormat, OutputFormat, and OutputWriter

Extracts InputFormat and OutputFormat enums (with parse() methods) and the
writeField CSV helper from main.zig into a new src/format.zig module.
Introduces OutputWriter, a stateful struct that dispatches output formatting
across all five formats (csv, tsv, json, ndjson, xml), eliminating the
format-switch inside execQuery. Closes no issue yet; part of #145.

* refactor: extract CLI argument types and parseArgs into args.zig

* refactor: consolidate SQLite helpers and ColumnType into sqlite.zig

* refactor: extract CSV loader and type inference into loader.zig

* refactor: extract runColumns, runValidate, runSample into src/modes/

* fix: address code review — consolidate ExitCode, remove dead code, guard XML validation

- Remove 4 identical copies of fatal() — all files now import the pub
  fn from sqlite.zig
- Collapse 19-case error switch in main() from ~100 lines to 22 using
  fatal() directly (it is noreturn)
- Replace if-chain parse() in InputFormat/OutputFormat with
  std.meta.stringToEnum — stays in sync with enum variants automatically
- Use beginTransaction/commitTransaction helpers in loader.zig instead
  of duplicating the inline sqlite3_exec logic
- Replace byte-by-byte takeByte() JSON reading in columns.zig and
  validate.zig with allocRemaining(.unlimited)
- Remove redundant @as(usize, N) casts in csv.zig unit tests

No behaviour change. All tests pass, ziglint clean.

* feat: add --disk flag for file-backed large dataset support

Switch from :memory: to a SQLite private temp file when --disk is set.
Also enables PRAGMA temp_store = FILE so ORDER BY/GROUP BY transient
structures spill to disk. Closes #91.

* fix: address review findings on --disk PR

- Update loop invariant comment to include disk flag
- Add explanatory comments on tests 125-127 (TEXT→NUMERIC SQLite conversion)
- Add tests 129-131: --disk + --no-type-inference, --header, --output

Updated version number in installation instructions for Debian, RPM, and APK.

* fix(xml): verify nested closing tag names in readContent

Maintain a tag stack inside XmlParser.readContent so that every
closing tag of a nested element is validated against the name of
the element that opened it.  Previously the name was read and
silently discarded, accepting malformed XML like
<col><a>text</b></col> without error.

Closes #134

* fix(xml): add safety comment on pop().? in readContent

… json.zig (#150)

Replace the tight takeByte() loops that read stdin one byte at a time
with reader.allocRemaining(allocator, .unlimited), which issues a single
bulk read using the Io vtable dispatch instead of one dispatch per byte.

Functions updated:
  - xml.zig: loadXmlInput, getXmlColumnNames, summarizeXml
  - json.zig: loadJsonArray

No API changes; all existing tests continue to pass.

Closes #135

…151)

* feat: add --json-path flag to navigate nested JSON arrays for input

Add --json-path <dot.separated.path> flag for -I json mode, allowing
users to reach a nested array inside a JSON document:

  curl https://api.example.com/feed     | sql-pipe -I json --json-path results.items 'SELECT title FROM t'

The flag is also forwarded to --columns and --validate so schema
inspection works on nested documents.  Omitting the flag preserves
existing behaviour (top-level bare array expected).

Adds navigateJsonPath() helper in json.zig, threads json_path through
ParsedArgs / ColumnsArgs / ValidateArgs, and adds 5 integration tests
(single-key path, multi-segment path, missing key, non-array target,
--columns with --json-path).

Closes #141

* fix: validate --json-path requires -I json; add validate test; renumber tests 132→133

Add --xml-root / --xml-row flag examples in all three documentation
locations so users can discover XML input support for nested documents
like RSS feeds:

- src/args.zig (--help): add XML example with --xml-root / --xml-row
- docs/sql-pipe.1.scd: add two XML EXAMPLES entries (nested input with
  --xml-root/--xml-row, custom output element names)
- README.md: expand XML section with RSS-style nested-input example

Closes #144