feat: rtk context — Show Claude Code context token budget

Use case

rtk context gives a clear picture of how many tokens my Claude Code setup costs per message, so I can decide if I need to slim down my CLAUDE.md or archive skills.

Scope MVP (v1)

Single command, zero flags. Scans and displays:

Target output

RTK Context Budget
--------------------------------------
Component                Tokens   Note
--------------------------------------
CLAUDE.md (global)        1,200
CLAUDE.md (project)         480
  @imports detected:
    @RTK.md                        not resolved (v1)
    @TONE.md                       not resolved (v1)
MEMORY.md (global)            0   not found
MEMORY.md (project)         187
Skills (47 loaded)        4,200
Commands (53 loaded)      2,150
Rules (1 file)              890
MCP servers (0)               0
--------------------------------------
TOTAL MEASURABLE          9,107
--------------------------------------
Note: System prompt + built-in tools add ~15K (not measurable by RTK)

Files to create/modify

Reuse existing

• tracking::estimate_tokens() (pub, src/tracking.rs)
• Table pattern from gain.rs (colored, TTY-aware, styled())
• dirs::home_dir() to locate ~/.claude/
• discover::provider::ClaudeProvider::encode_project_path() to find the project folder

Error handling

• ~/.claude/ absent → "Claude Code not detected"
• Missing file → skip, display 0 + "not found"
• settings.json absent/malformed → display "unknown"

Tests

• Inline #[cfg(test)] with tempdir
• Test detect_imports() on @file.md patterns
• Test missing component → 0 tokens
• Snapshot test on output format

Verification

cargo fmt --all && cargo clippy --all-targets && cargo test
./target/debug/rtk context   # on real setup
rtk --help | grep context
hyperfine 'rtk context'      # target <100ms

Out of scope (v2, defer)

• --format json
• --project scope
• Recursive @import resolution
• --detailed per-file breakdown

Acceptance criteria

• rtk context displays a table with all listed components
• Each token count is based on real file content (no fixed estimates except MCP)
• Missing file = graceful skip (no panic)
• cargo test passes, clippy clean
• CHANGELOG.md + README Module Responsibilities table updated

Start with src/main.rs for command definitions and routing, then read src/tracking.rs, gain.rs, and discover/provider.rs for token estimation, table output, and project-path encoding. Create src/context_cmd.rs and update scripts/test-all.sh, covering the listed components, missing-file handling, inline tests, and output snapshot. Run cargo fmt, clippy, cargo test, and the documented context smoke checks; done includes the README and CHANGELOG updates.