Phase 4 — Cutover: make joint-descent the default

[k1sauce]

Part of #1. Depends on #5.

Scope

Once chr22 MAPQ ≥99% via the full-descent path, promote it to default. The current independent-then-rescue path becomes the legacy fallback.

Deliverables

• Flip default: `Cli::joint_descent: bool` defaults to `true`. Add `--legacy-paired` opt-out flag.
• Remove or hide the env-var tuning knobs (`RUSTTIE_MAX_RESCUE`, etc.) — settled defaults baked in.
• Update `README.md` validation table with new MAPQ numbers (target 99%+).
• Update the "Known MAPQ gap" section in `README.md` — either delete it (gap closed) or document the remaining minor cases.
• Add a regression test in `crates/rusttie-cli/tests/` that asserts MAPQ agreement floor on the lambda-virus mini-corpus (CI gate).
• Cleanup: remove the bounded-Cartesian Cartesian fallback in `classify_pair_set` if no longer reachable.

Exit criterion

• New default produces ≥99% MAPQ on chr22 in CI.
• Default-path wall time within 2× of current baseline.
• All existing tests pass.

Risks

• Making joint-descent the default changes wall time (~3× of current baseline). Some users may prefer the faster legacy path; `--legacy-paired` keeps that available.
• The legacy path becomes dead-but-not-dead code. Need clear deprecation messaging.