perf(cli): cut per-cycle invocations — merge render-only calls into fetch, single entry point

[OrenAshkenazy]

Problem

A normal cycle costs ~8-10 separate python invocations: root-resolution, fetch, --profile-intro, --planned-verification, run_profile.py, --cycle-summary, request_rereview.py, --wait, --record-run. Each is a full agent turn plus ~120 chars of "$GGRL_PLUGIN_ROOT/skills/gh-review-loop/scripts/..." path boilerplate per command.

Proposal

• Fetch already knows profile state — append the profile-intro and planned-verification blocks to fetch output. Kills 2 calls per cycle.
• Single ggrl <subcommand> entry point (shim or python3 -m) to kill path boilerplate.

Tradeoffs

• Merged outputs are bigger and less composable for debugging.
• A shim adds install/version-skew complexity across the plugin-cache copy mechanism.

[OrenAshkenazy]

Open-issue / PR audit

Status: partially solved, despite PR #107 saying Closes #99.

PR #107 implemented the first proposal exactly: --profile-intro and --planned-verification are now carried in the normal fetch output, removing two per-cycle invocations. Because #107 was stacked, PR #108 is the PR that ultimately landed that combined work on main.

The second proposal is still missing: there is no single ggrl <subcommand> / python3 -m ... entry point, and the workflow still invokes separate scripts with long plugin-cache paths.

Conclusion: #107 (landed via #108) solves half of #99, not the whole issue. Keep #99 open, or split the remaining single-entry-point work into a follow-up issue and close this one as partially delivered.