perf(ruby): remove two superlinear costs from the RuboCop pass

[amritk]

What & why

Two performance patches for the RuboCop pass, applied at boot the way the existing syntax_tree patches are, so the artifact stays stock. Neither changes a byte of output; both change what a gem costs.

1. The token sort, on every file with a heredoc in it. rubocop-ast's ProcessedSource#sorted_tokens falls back to a stable sort keyed on a two-element Array, [begin_pos, index], whenever the token list arrives out of position order — which any heredoc causes. sort_by compares Integers inline and Arrays through a method call into Array#<=>: ~150 ms on a 49 KB file with 4,738 tokens. The patch folds the pair into one Integer, begin_pos * count + index, which orders identically and compares as an Integer: ~9 ms. Over a 545 KB profile of 60 real files the sort had been 16.5% of everything the formatter did. (rubocop-ast master has since made the same change, unreleased; the patch retires when the pin moves past 1.50.0.)

2. The line table, on every file with a multi-byte character in it. The same bug the 0.6.2 syntax_tree patch fixed, one layer down. parser's Buffer#line_begins walks the source with String#index("\n", from), which counts characters from the start on a non-ASCII string, so the table is O(size × lines) to build — and RuboCop rebuilds it on every correction round. On a 589 KB file with 17,401 lines and 729 carrying a multi-byte character it took 9.0 s per round. The patch walks the source once with each_line; the same table builds in ~25 ms. An ASCII source keeps the gem's own loop.

Node, one process, gems' own methods restored for the first run	before	after
206 files, 2.0 MB (rubygems, bundler, stdlib; 135 with a heredoc, 4 non-ASCII)	120.5 s	79.2 s
the 589 KB file alone	52.1 s	27.4 s

Which packages

packages/ruby

Exactness

Output is unchanged, and three things say so:

• src/rubocop-perf-patch.test.ts asserts the patched sort returns the tokens object for object in the order the gem's own expression puts them, on five heredoc shapes that each reach the sort; and that the patched line table matches the gem's own loop entry for entry, sentinel included, on thirteen shapes of source (CRLF endings and multi-byte characters among them). A further test asserts both methods are the patched ones, so a patch that silently stops being applied fails rather than passes.
• test/rubocop-conformance.test.ts gains one sample per patch, run through the real rubocop binary on a native Ruby with the pinned gems. It ran here (not skipped) and passes.
• bun run ruby:bench --only corpus over the 206-file corpus above, snapshotted before and after and --compared: 206 of 206 hashes the same.

How it was validated

• bun run check
• bun run types:check
• bun run test
• bun run test:node
• Rebuilt the wasm artifact — not needed; the patches are applied at boot and the artifact is untouched

Changeset

.changeset/quick-heredoc-token-sort.md, a patch bump for @scalar/ruby-fmt.