feat: per-provider concurrency lanes (cloud parallel while local churns)

entrypoint.sh groups models by provider into lanes that run in PARALLEL; within
a lane at most `cap` models run at once. cap = GADFLY_PROVIDER_CONCURRENCY map
("ollama-cloud=3,m1pro=1") else GADFLY_CONCURRENCY (default 1). So a single
local box stays serial (1 at a time) while cloud models run several at once and
both lanes progress simultaneously. Portable bash (no associative arrays).
Default cap 1 keeps a single-provider pool sequential as before. Pairs with the
per-lens timeout so a slow lane can't starve others. Docs: README Concurrency
section + config table; CLAUDE.md lessons incl. the docker://:latest cache gotcha.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

CLAUDE.md

@@ -127,3 +127,11 @@ are actually exercised. OpenAI/Anthropic/Google come from majordomo's abstractio
   `GADFLY_ENDPOINT_<NAME>="provider|base-url[|key]"` mechanism (see `cmd/gadfly/model.go`).
 - Gitea `vars`/`secrets` are **not** auto-exposed as env in a job — the consumer stub must map
   each one explicitly in its `env:` block (dynamic alias names can't be auto-enumerated).
+- **`uses: docker://…:latest` is CACHED by act_runner** — a freshly-pushed `:latest` is often
+  NOT re-pulled, so the job silently runs the previous image. For a run that must use a specific
+  build (e.g. validating a just-pushed fix), pin the consumer stub to the immutable
+  `:sha-<short>` tag the build publishes, not `:latest`.
+- **Concurrency is per-provider** (`entrypoint.sh`): each provider is a lane, lanes run in
+  parallel, `cap` (from `GADFLY_PROVIDER_CONCURRENCY` else `GADFLY_CONCURRENCY`, default 1) bounds
+  models-at-once within a lane. The review timeout (`GADFLY_TIMEOUT_SECS`) is **per-lens**, not
+  shared across the suite — a slow model can't starve later lenses (the original timeout bug).