refactor(cli): split main.rs into per-responsibility modules

[thiagomargatho]

Summary

• main.rs had grown to ~15k lines with 8+ unrelated concerns interleaved function-by-function throughout the file (arg parsing, boot/health checks, runtime bootstrap, provider client, progress reporting, model resolution, skill dispatch). Split along those natural boundaries into separate files; main.rs is now a thin entry point (constants, main(), mod declarations, existing tests).
• Also fixes a stale path in root CLAUDE.md (.claude/settings.local.json → .claw/settings.local.json, matching rust/CLAUDE.md and .claw.json).
• No logic changed — items moved verbatim, marked pub(crate) where a new module boundary required it, with use crate::*; in each new file so cross-module references resolve through main.rs's re-exports.

New files:

• model_provenance.rs — model alias/version resolution
• cli_parse.rs — CLI arg parsing, help, report formatting
• preflight.rs — doctor/health checks, boot snapshot, git status
• bootstrap.rs — runtime bootstrap, LiveCli, session management
• provider_client.rs — Anthropic client, tool executor, MCP wiring
• progress.rs — prompt/hook progress reporting
• skill_dispatch.rs — bare-skill dispatch, cwd guard, diff reports

Test plan

• cargo clippy --workspace --all-targets -- -D warnings — clean
• cargo test --workspace — all suites pass
• scripts/fmt.sh --check — clean

🤖 Generated with Claude Code

[1716775457damn]

Nice refactor! The decomposition into per-responsibility modules is clean and well thought out. A couple of observations:

• Each new file using use crate::*; to resolve through main.rs's re-exports works but creates a tight coupling — consider whether some of these modules could eventually be promoted to first-class crate modules with their own explicit imports, which would make dependencies more explicit and testable in isolation.
• model_provenance.rs feels like it could be simplified to just provenance.rs or resolution.rs — the "model_" prefix is redundant given the file is already in a CLI context.
• The .claw path fix in root CLAUDE.md is a good catch.

Test plan looks solid — clippy clean, all tests pass, fmt clean. No blockers from my side.