fix: prevent rtk read from corrupting JSON/data files (#464)

[ousamabenyounes]

Summary

Fixes #464 — rtk read package.json corrupts JSON files when string values contain /* or */.

Root cause: .json files were classified as Language::Unknown, which uses /* / */ as block comment delimiters. The string "packages/*" was interpreted as opening a block comment, and "**/package.json" as closing it — everything between was silently deleted.

Fix: Add Language::Data variant with no comment patterns. JSON, YAML, TOML, XML, CSV, Markdown, and other data formats skip all comment stripping and code filtering entirely.

Before/After

# BEFORE: rtk read corrupts the JSON structure
$ rtk read package.json
{
  "name": "my-monorepo",
  "workspaces": {
    "packages": [
      "sort-package-json",                    # <-- from lint-staged!
      "biome check --write --no-errors-on-unmatched"
    ]
  }
}
# scripts: MISSING, lint-staged: MISSING, catalog: MISSING

# AFTER: JSON structure fully preserved
$ rtk read package.json
{
  "name": "my-monorepo",
  "workspaces": {
    "packages": [
      "packages/*"                            # <-- correct
    ],
    "catalog": { ... }
  },
  "scripts": {                                # <-- preserved
    "build": "bun run --workspaces build",
    "lint": "bun run --workspaces lint"
  },
  "lint-staged": {                            # <-- preserved
    "**/package.json": [ ... ]
  }
}

Why this matters

package.json is read dozens of times per Claude session. When corrupted:

• Claude sees wrong workspace config, missing scripts, broken lint rules
• Makes decisions based on false metadata (wrong build commands, missing dependencies)
• Re-reads the file in a loop trying to reconcile the corruption = wasted tokens

Also affects: tsconfig.json, docker-compose.yml, Cargo.lock, .env, schema.graphql, *.sql, etc.

Affected extensions

json, jsonc, json5, yaml, yml, toml, xml, csv, tsv, graphql, gql, sql, md, markdown, txt, env, lock

Test plan

• test_language_detection_data_formats — all data extensions map to Language::Data
• test_json_no_comment_stripping — reproduces exact bug: rtk read corrupts package.json when JSON strings contain /* or */ #464 scenario (packages/*, scripts, lint-staged)
• test_json_aggressive_filter_preserves_structure — aggressive filter also safe for JSON
• Full suite: 767 passed, 0 failed
• Manual test: rtk read package.json with the exact fixture from the issue
• validate-docs.sh passes

Generated with Claude Code

[pszymkowiak]

LGTM — good fix for a nasty bug. Adding Language::Data with empty comment patterns is the right approach.

Build passes, 55 filter tests pass. The exact reproduction case from #464 (packages/* treated as block comment) is covered.

One nit (non-blocking): the version bumps in ARCHITECTURE.md/CLAUDE.md/README.md are unrelated noise — ideally drop them from this PR since release-please handles versions. But not a blocker.

@aeppling for merge.

[ousamabenyounes]

Thanks! Removed the version bumps from ARCHITECTURE.md, CLAUDE.md and README.md — rebased on latest develop first, so the PR now only touches src/filter.rs (1 file, 42+/10-).

Note: the Benchmark and Documentation Validation checks are failing due to pre-existing issues on develop (missing truncate function in cargo_cmd.rs:838, and version string mismatch in docs). Not related to this PR.