Template aware parsing (#143)

* Add language-aware class sorting

Keep template expressions opaque while sorting adjacent static runs.

Infer profiles from paths and preserve legacy behavior for unknown sources.

* Fix language-aware class extraction regressions

Preserve template expressions across inferred and explicit language
profiles while keeping valid Tailwind punctuation sortable in unknown
files.

Use indexed markup spans for large documents, respect Svelte component
case semantics, and restore positional custom-regex captures.

* Document language-aware sorting changes

Record the new language profiles, template-safety fixes, compatibility
behavior, performance work, and public API migrations.

* Update cache and string dependencies

* Bump Rust deps and add update task

Upgrade several crate dependencies across `rustywind-core` and `xtask`, including benchmarking, allocator, CLI, networking, and archive tooling. Also add a `just update` recipe to refresh workspace dependencies more easily.

* Adapt code to updated dependencies

* Parse Astro and templates with Winnow

Add typed markup extraction and first-class Astro handling while
preserving template islands. Avoid repeated tag scans and keep
legacy-compatible Lit and Ruby markup sortable.

* Document Astro parsing and breaking changes

* Add short CLI aliases

Add single-letter aliases for common options to reduce command input
cost while keeping full-length flags supported.

Update README usage output, changelog notes, and parsing tests to cover
the new -p, -l, and -f behavior.

* Prevent class sorting from rewriting templates

Validate wrapped captures and attribute names before sorting. Respect
HTML casing, PHP comment delimiters, and escaped Blade directives so
source expressions remain byte-exact.

* Parse Svelte block tags for class sorting

Separate document template islands from class-value expressions so Svelte
closing tags no longer disable sorting for the entire file.

Track Svelte block structure and preserve fail-closed behavior for malformed
markup.

Author: praveenperera

CHANGELOG.md

@@ -2,6 +2,49 @@
 
 ## [Unreleased]
 
+### Added
+
+- Add language-aware class extraction for HTML, Svelte, Django, Jinja, Twig,
+  Liquid, Handlebars, ERB, EJS, PHP, Blade, Lit, and Ruby
+- Add `--language` to override source-language inference and
+  `--stdin-filename` to infer the language of standard input
+- Preserve template expressions as opaque boundaries while independently
+  sorting adjacent static class runs
+
+### Fixed
+
+- Preserve Django and Jinja output expressions and control tags, including
+  expressions attached to class names, [#138](https://github.com/avencera/rustywind/issues/138)
+- Prevent ERB ternaries and quoted expressions from being corrupted,
+  [#140](https://github.com/avencera/rustywind/issues/140)
+- Prevent Svelte inline conditionals from being reordered as static classes,
+  [#142](https://github.com/avencera/rustywind/issues/142)
+- Keep ambiguous template expressions in plain `.html` files unchanged unless
+  an explicit template-language profile is selected
+- Sort valid Tailwind punctuation and quoted arbitrary values in files with
+  unrecognized extensions while continuing to reject template expressions
+- Preserve nested class attributes inside capitalized Svelte components
+- Continue accepting the first positional capture in `--custom-regex`, as in
+  RustyWind 0.24 and earlier, while also supporting a named `classes` capture
+
+### Performance
+
+- Index markup tag spans so class extraction no longer scans every tag for
+  every class attribute in large documents
+
+### Breaking changes
+
+- `RustyWind::has_classes` now accepts a `SourceDocument`,
+  `RustyWind::sort_file_contents` is replaced by `sort_document`, and
+  `RustyWind::sort_classes` is replaced by `sort_class_list`, which requires a
+  validated `PlainClassList`
+- `FinderRegex::CustomRegex` now contains a `CustomClassExtractor`; library
+  callers constructing this variant must validate their `Regex` with
+  `CustomClassExtractor::new`
+- Recognized markup languages only sort literal `class` and `className`
+  attributes in actual tags. Class-like text in comments, raw-text elements,
+  program strings, and dynamic or namespaced attributes is no longer rewritten
+
 ## [0.25.2] - 2026-07-08
 
 ### Fixed