feat(fs): auto-compress cat'd markdown via compress_md pipeline

[claudioemmanuel]

Summary

• Added extract_md_viewer_target() helper in src/commands/fs.rs that detects viewer commands (cat/head/tail/less/more/bat) targeting .md files
• New md-mode branch in FsHandler::compress routes .md files through compress_md::compress_text_with_locale when auto_compress_md=true (the default), bypassing smart_filter/dedup/truncation entirely
• Emits [squeez: md-mode — compressed N→M tokens from path] header; falls back to raw text if integrity check fails
• Adds 3 tests to tests/test_fs.rs covering the happy path, disabled flag, and non-.md files
• Adds bench/fixtures/cat_claude_md.txt (486 lines) picked up by bench/run.sh

Motivation

Large plan files like plan.md at 1,754 lines were being read repeatedly with no compression — the existing FsHandler only head-truncated to 50 lines. The markdown compressor was already proven on memory files but unused for ad-hoc cat invocations. This change closes that gap with zero new dependencies.

Benchmark impact

FIXTURE                               BEFORE    AFTER  REDUCTION  LATENCY STATUS
cat_claude_md.txt                     18406tk     437tk        98%      12ms  ✅

98% token reduction on a 486-line representative CLAUDE.md fixture.

Test plan

• cargo test --test test_fs — 5 tests pass (2 existing + 3 new)
• cargo test — full suite green
• bash bench/run.sh — new fixture shows 98% reduction
• cargo build — zero warnings