feat(ignore-directives): support block-scoped deno-lint-ignore

[bartlomieju]

A // deno-lint-ignore directive only suppressed diagnostics on the
single line that follows it, so silencing a rule across a multi-line
span meant repeating the directive on every line or reaching for the
much coarser deno-lint-ignore-file.

A directive placed directly above a { ... } block now suppresses
diagnostics anywhere inside that block. Coverage is intentionally
limited to bare, explicit blocks: the directive has to be a leading
comment of a BlockStmt. A directive placed before a function, class
or other statement attaches to that node rather than to a block, so it
keeps the ordinary next-line behavior and does not silently swallow
deeply nested diagnostics, which was the main concern raised when this
was discussed.

Closes #476

[bartlomieju]

Verdict: LGTM — correct and well-tested, including the key safety guard. Recommend merge (ignore-directive behavior change → maintainer sign-off; needs green CI / rebase).

Correctness — verified the main risk

• The over-suppression concern (a directive before if (x) {/function f() { swallowing the whole body) is handled correctly and tested: because comments attach to the nearest following token, leading_comments(block_range.start) only returns a directive that sits directly above a bare {. The test function foo(): any { let a: any } yields block_end_line() == None, and the behavioral test confirms debugger inside a function body is still reported. So block scope only applies to explicit bare blocks — exactly the intent.
• The containment check (block_start_line <= diagnostic_line <= block_end_line + check_used) is right, and it runs only after the ordinary next-line path, so existing behavior is unchanged when no bare block is involved.
• extend_block_ignore_directives is correctly gated on !directives.is_empty(), so files without directives pay nothing.

Performance — minor

• The block-directive check in the diagnostic filter is an O(block-directives) linear scan per diagnostic, i.e. O(diagnostics x block-directives) worst case. Negligible for normal files (few directives), but if you expect files with many ignore directives + many diagnostics, consider indexing block directives by line range. Not a blocker.
• extend_block_ignore_directives walks the full AST once per file-with-directives — acceptable as a one-time per-file cost.

Suggestion

• Add an if/for/while body case to the "blocks only" tests (function is covered; the attachment mechanism is the same, but explicit coverage would lock in that if (x) { bodies aren't block-scoped).

Housekeeping: mergeable: UNKNOWN → rebase. Behavior change to ignore-directives → your call to sign off (your own PR).

No changes requested. Well-scoped feature with the right guardrails and tests; good to merge once CI is green.