feat: deep token reduction — 4 phases of output compression improvements

[ToyStoryCoder]

Summary

4 phases of improvements to reduce LLM token consumption across all RTK command handlers.

Phase 1 — Stability & Code Quality

• Replace unwrap() calls with proper error handling
• Eliminate code duplication across command handlers
• Add supply chain security with deny.toml (cargo-deny)
• CI workflow improvements

Phase 2 — Core Token Savings

• Lockfile detection (is_generated_file) for snapshots, minified assets, codegen
• SmartFilter level with import/test collapsing, long string truncation
• JSON array truncation (>50 items) with summary
• Grep output deduplication for repeated patterns
• Similar-file grouping in find output

Phase 3 — Extended Filters

• Tree output cap (MAX_TREE_LINES = 200)
• Git show --stat filtering (trim long paths, cap file list)
• Git log width cap (120 chars)
• Find filename grouping by extension
• JSON support for Smart filter level
• 3 new TOML filters (docker-build, flutter-build, pip-install)
• Grep --no-compact flag support

Phase 4 — Deep Token Reduction

• NOISE_DIRS expansion (+11 entries: .gradle, .cargo, .npm, vendor, etc.)
• Git branch remote cap reduced 10→5
• Stash list cap (MAX_STASH_ENTRIES = 20)
• SmartFilter blank line normalization (3+ → 2)
• SmartFilter doc-comment collapse (>15 lines → first 3 + summary)
• Lockfile summary in rtk read (14 lockfile names)
• ls entry cap (MAX_LS_ENTRIES = 50)
• Curl HTML tag stripping (script/style removal, entity decoding)

Test plan

• 1283 tests passing, 0 failures
• No new clippy warnings introduced
• 24 files changed, +1542 -111
• All filter levels preserve escape hatch (--filter none / -l none)

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}