feat(no-empty-file): add no-empty-file rule

[bartlomieju]

Ports oxlint's unicorn/no-empty-file rule to deno_lint as a native Rust
rule. It flags a file whose top-level body contains no real code — only
whitespace, comments, directive prologues (e.g. "use strict";), empty
statements (;), or empty blocks ({}).

Top-level string-literal statements are treated as directives and counted
as empty, matching oxc; a string literal nested inside a block is a real
expression statement and keeps the file non-empty. Files consisting solely
of a triple-slash reference directive are exempt. The reported span is
capped at 100 characters so comment-heavy empty files don't produce a huge
diagnostic.

Reference rule:
https://github.com/oxc-project/oxc/blob/main/crates/oxc_linter/src/rules/unicorn/no_empty_file.rs
(test fixtures ported from there; eslint-disable-specific fixtures were
dropped as they don't translate to deno_lint's ignore mechanism).

Note for maintainer review: this rule is tagged RECOMMENDED (on by
default for every user), consistent with it being a correctness rule.
Because of that, an empty file now produces a diagnostic by default, which
required updating the existing empty_file and empty_file_with_ast
framework tests in src/lib.rs to expect the new no-empty-file
diagnostic. Please confirm the on-by-default tagging is desired.

[bartlomieju]

Verdict: LGTM on the logic — needs maintainer sign-off (RECOMMENDED/on-by-default) + one perf reorder. Not auto-merging (new rule).

Faithful port of oxlint's unicorn/no-empty-file.

Correctness — verified

• Top-level directive strings ("use strict";) count as empty; a string nested in a block is real code — the is_empty_top_level_stmt vs is_empty_nested_stmt split handles this correctly, matching oxc.
• Stmt::Expr only treats Expr::Lit(Lit::Str) as a directive, so a top-level template literal `x`; is (correctly) real code, not a directive. Nice distinction.
• ModuleDecl (imports/exports) → non-empty; vacuous .all() on an empty body → flagged. Correct.
• The 100-char span cap uses char_indices().nth(100) → a byte offset at a char boundary, so start + byte_offset is UTF-8 safe. Good.

Performance — worth reordering

• The triple-slash exemption context.all_comments().any(is_triple_slash_reference) runs first, for every file, scanning all comments and regex-matching each line comment — even for obviously non-empty files. Since body-emptiness is cheap (just iterate top-level items) and almost always false, check emptiness first and only do the comment scan when the file otherwise looks empty:

  let is_empty = match program { ... };
  if !is_empty { return; }
  if context.all_comments().any(is_triple_slash_reference) { return; }

  That moves the per-file comment scan off the hot path.

Minor

• The triple-slash regex char class ["|'] includes a literal | as an alternative quote char (should be ["']). Harmless for real /// <reference> directives, just a smell.

Product decision (you flagged it)

• Tagged RECOMMENDED, so empty files now warn by default for everyone (hence the src/lib.rs framework-test updates). Reasonable for a correctness rule, but it's the call to confirm — some users intentionally keep empty mod.ts/placeholder files.

Housekeeping: mergeable: UNKNOWN → rebase; docs .md? (not in the diff — new rules conventionally ship one).

Logic is solid and well-tested. Blockers: confirm the RECOMMENDED tagging + the perf reorder; then good to land.