feat(check): retire the thinking-disabled workaround — effort buys extended thinking

The cersei pin predated the provider's ability to round-trip Anthropic
thinking blocks: its SSE parser dropped `signature_delta`, so the thinking
block echoed back on the second turn carried an empty signature and the API
rejected it. Effort was therefore applied as a sampling temperature, with a
comment promising to revisit once fixed upstream.

wack/cersei (now the authoritative source; we are not tracking pacifio
upstream) carries the fixes: 94f18b2 accumulates the signature from
`signature_delta` and 5bd06db preserves `redacted_thinking` across turns.
Bump the pin (d485d50 -> 4dfac7f — which also brings the bounded,
gitignore-aware GlobTool that closes the unbounded-walk timeout vector)
and map effort onto its natural vehicle:

- medium/high enable extended thinking (4096/8192-token budgets, cersei's
  own EffortLevel scale, safely under the 16k default max_tokens);
- low — the default — keeps thinking off for speed and cost, steering with
  temperature exactly as before (0.0, +0.5 per retry attempt).

Exactly one control is applied per run: the API rejects a temperature when
thinking is enabled (thinking samples at 1.0, which also gives thinking
retries the trajectory diversity that attempt_temperature provides on
thinking-free runs).

Not yet verified against the live API (no credentials in the build
environment): a `multi check --effort medium` run on a real project should
confirm turn-2 thinking blocks are accepted before this merges.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VxKv1hhPZ4GocfmwHUk1G8

Author: RobbieMcKinstry

BUGS.md

@@ -2,20 +2,21 @@
 
 These are bugs (or missing features) I've observed while working with `multi checks`.
 
-- [ ] Output is now hanging. I suspect this is recent (within the last few commits) and it started
-happening after implement the changes to the `Presenter` actor to fix writing text off-screen without wrapping.
-
 - [ ] Remove the `Claude -p` executor.
 
-- [ ] Running 16 agents seems to nearly freeze the computer. Use an OTel profile to determine if this is true.
+- [ ] No use of Cersei workflows to chain multiple prompts together.
 
-- [ ] Temperature not configured.
+- [ ] Logs no longer report the id of the check that failed (or the number of attempted retries)
 
-- [ ] No limit on max turns.
+- [ ] Assemble_instructions is hard-coded: src/checks/executor/mod.rs:98 (definition), called from src/checks/executor/cersei.rs:110
 
-- [ ] Logs no longer report the id of the check that failed (or the number of attempted retries)
+- [ ] No system prompt provided.
 
-- [ ] No use of Cersei workflows to chain multiple prompts together.
+- [ ] Not sure if prompt caching is enabled at all.
+
+- [ ] Running 16 agents seems to nearly freeze the computer. Use an OTel profile to determine if this is true.
+
+- [ ] No limit on max turns.
 
 - [ ] No support for Fireworks AI.
 
@@ -25,14 +26,6 @@ happening after implement the changes to the `Presenter` actor to fix writing te
 
 - [ ] No loading of RULES.md files from the .claude directory.
 
-- [ ] Assemble_instructions is hard-coded: src/checks/executor/mod.rs:98 (definition), called from src/checks/executor/cersei.rs:110
-
-- [ ] No system prompt provided.
-
-- [ ] Not sure if prompt caching is enabled at all.
-
-- [ ] No trace capture. We need a way to record all session traces so that we can analyze why they failed.
-
 - CERSEI: `append_system_prompt()` function is dead unless routed through the separate build_system_prompt() composer.
 
 - [ ] `Ctrl-C` (shutdown signals) needs to be handled gracefully and cross-platform.
@@ -50,6 +43,13 @@ guaranteeing the terminal is restored on the way out.
 
 ## Fixes
 
+- [x] No trace capture. We need a way to record all session traces so that we can analyze why they failed.
+
+- [x] Output is now hanging. I suspect this is recent (within the last few commits) and it started
+happening after implement the changes to the `Presenter` actor to fix writing text off-screen without wrapping.
+
+- [x] Temperature not configured.
+
 - [x] No loading of CLAUDE.md files
 
 - [x] Concurrency still not respected.

Cargo.lock

@@ -187,9 +187,10 @@ base_url = "https://..."
 An unset flag contributes nothing — it never overrides a value from the
 environment or file. The `model` is validated against a hardcoded allowlist of
 known IDs for the selected provider; an unknown ID is a clear error. `effort`
-currently maps to the in-process agent's sampling temperature (`low` → most
-deterministic, `high` → most exploratory); mapping it to an extended-thinking
-budget is pending an upstream provider fix.
+maps to the in-process agent's extended-thinking budget: `medium` and `high`
+enable extended thinking (4096- and 8192-token budgets respectively), while
+`low` — the default — keeps thinking off for speed and cost, running the agent
+deterministically instead.
 
 The **`executor`** selects the execution engine. The default `cersei` runs each
 check as an in-process agent (native multi-provider model swapping, no external