fix(tickets): continue compound reads after retrieval

[BjRo]

Why

Closes #208. read-ticket ended compound requests after retrieval, so users did not receive the separately requested research or analysis.

What changed

Updated TM-R3/TM-R4 and read-ticket so standalone reads relay the CLI stream unchanged while compound reads retain the complete stream as evidence and continue according to the user's request. Retrieval failures stop dependent work and allow independently meaningful work to continue with the complete diagnostic. Clarified the bundled launcher path, added compound evals, and bumped darrow-tickets to 0.5.2.

Verification

• bun evals/runner/run.ts --skill read-ticket --harness codex --model gpt-6-luna --effort medium --trials 5 --jobs 3 --no-progress: 16/16 cases met the 80% threshold; 79/80 individual trials passed.
• bun run lint, bun run lint:ts, bun run lint:shell, bun run typecheck, bun run check:decisions, and bun run check:docs: passed.
• The skill inspector and git diff --check: passed.
• A focused pre-fix Codex trial reproduced the missing follow-on recommendation. This was exploratory rather than a matched n=5 control.

Review notes

One compound-success trial copied an invalid absolute launcher path and stopped before retrieval; that case passed 4/5. No bundled backend code changed. Review the standalone exact-output branch and the compound failure continuation rules.

Checklist

• I have read and followed CONTRIBUTING.md, including the contribution licensing terms.
• I added or updated the applicable invariant before implementation, or this change does not affect a capability invariant.
• I added or updated colocated evals, or this change does not affect skill behavior.
• I confirmed that each changed plugin remains self-contained, or this change does not affect plugin content.
• I ran bun run check:python, or this change does not affect registered Python packages or their repository quality infrastructure.