Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@crates/biome_module_graph/README.md`:
- Line 3: Fix the grammar in the README sentence that currently reads "A module
a is a known file that imports and exports data, which is used" by removing the
extra "a" so it reads "A module is a known file that imports and exports data,
which is used"; update the sentence in the README.md (the line containing "A
module a is...") to the corrected wording.

In `@crates/biome_module_graph/src/css_module_info/traverse.rs`:
- Around line 105-143: The loop over importers currently breaks after the first
CSS-producing parent, which prevents pushing later sibling file_path entries
onto self.stack; change the logic in traverse.rs so you do not exit the loop
early: instead of break, continue iterating so all sibling importers get pushed,
and only set self.current_css_iter if it is not already set (guard with
self.current_css_iter.is_none()) to avoid overwriting a previously chosen
iterator; keep the existing pushes to self.visited and self.stack and the
css_steps creation but replace the unconditional break with a continue and a
conditional assignment for current_css_iter.

In `@crates/biome_module_graph/src/db/queries.rs`:
- Around line 392-397: The default-reexport branch (ImportSymbol::Default) must
be guarded by seen_paths just like the Named branch to avoid infinite loops:
when you resolve reexport.import.resolved_path.as_deref() and call
db.js_module_info_for_path(path) before doing stack.push((module, symbol_name)),
check whether seen_paths already contains that path and only push (and/or insert
into seen_paths) if it is not present; apply this exact same guard in both
functions that handle exported-symbol lookup and JSDoc lookup (the other
occurrence around the second ImportSymbol::Default handling) so default
re-export cycles are short-circuited.
- Around line 180-204: The code iterates html_info.static_import_paths twice
causing duplicate CssClassStep entries; update the second loop to only iterate
dynamic imports (or otherwise exclude static_import_paths) so each import is
processed once: modify the loop that currently chains
html_info.static_import_paths.values().chain(html_info.dynamic_import_paths.values())
to only iterate html_info.dynamic_import_paths.values() and keep the same
css_module_info_for_path lookup and linked_steps.push(CssClassStep { css_path:
..., css_classes: ... }) logic to avoid duplicated CSS steps.

In `@crates/biome_module_graph/src/path_info_cache.rs`:
- Around line 50-53: The loop in while let Some(dir) = parent currently calls
self.get_or_insert(dir, fs) and breaks if that returns Some, which treats
newly-inserted entries the same as pre-existing ones and stops prepopulating
ancestors; change the logic so you first check existence with self.get(dir) (or
an existence check) and only break when the directory already exists, otherwise
call self.get_or_insert(dir, fs) to insert and continue up the ancestor chain;
update get_or_insert only if needed (or have it return a flag) so the loop can
distinguish newly-inserted vs pre-existing entries.

In `@crates/biome_service/src/workspace/server.rs`:
- Around line 1435-1473: These methods silently ignore a poisoned DB mutex by
returning on lock error; instead panic immediately so callers don't operate on
stale state. Replace the pattern "let Ok(...)= self.lock_db() else { return; }"
in db_set_module_info, db_remove_module, and db_unload_path with a clear
unwrap/expect from lock_db() (e.g. let mut db =
self.lock_db().expect("WorkspaceServer db lock poisoned") for the mutable case)
so the code fails fast on poison rather than silently no-op.

---

Outside diff comments:
In `@crates/biome_js_analyze/src/lint/correctness/no_unresolved_imports.rs`:
- Around line 137-145: The lint is calling
has_exported_symbol/find_js_exported_symbol for any ModuleInfo; add a guard to
only run export checks when target_info.kind is ModuleInfoKind::Js: after
obtaining target_info via ctx.module_info_for_path(resolved_path) (and before
constructing GetUnresolvedImportsOptions and before the has_exported_symbol
path), check target_info.kind == ModuleInfoKind::Js and skip export-resolution
logic (return Vec::new() or bypass has_exported_symbol) for non-JS kinds; ensure
both the wiring around target_info (where GetUnresolvedImportsOptions is
created) and the branch that calls has_exported_symbol/find_js_exported_symbol
(the code between the existing lines ~239–275) are updated to short-circuit for
non-JS ModuleInfoKind.

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@crates/biome_css_analyze/src/lint/nursery/no_unused_classes.rs`:
- Around line 78-85: The new DB-backed lookup path using ctx.db(),
module_for_path(...), is_class_referenced_by_importers(...) and SymbolName built
from class_name.token_text_trimmed().text() needs corresponding spec fixtures:
add explicit valid and invalid test cases for the noUnusedClasses rule that
exercise importer-resolution (a class referenced only via an importer should be
marked valid) and the :global(...) exclusion (classes inside :global(...) should
be treated as used/valid). Create fixtures that construct SymbolName via the
same token text shape and include a negative case where a class is truly unused
to assert the rule flags it.