fix(agent): remove the todo-stall pause

[esengine]

The last remaining way for the host to end a turn the user never asked it to end.

What it was

A stalled todo could end a run: 8 rounds without a new completion, read, command or mutation got a reassessment nudge, 16 got a resumable pause. The pause goes; the nudge stays.

It was not the mechanism that caught what it claimed

The zero-evidence ladder escalates on the same receipts and reaches its stop at 6; the storm breaker reaches its own at 3. The todo streak reached 16.

Writing the Goal test for this proved it outright: under Goal the run never got to 16 at all, because the ladder's structural pause fired first every time. So this was dead code under Goal, and in ordinary chat it was the only host-owned stop left after #8320 and #8340.

What it measured was also narrower than it reads. Any first-time successful read, command or mutation renews the streak - progressReceiptSignature counts reads, and folds the output digest into the signature, so even re-reading a file whose contents changed renews it. Sixteen rounds without renewal means sixteen rounds producing literally nothing new: exactly the state the two guards above already handle, twice over and far sooner.

What is removed

• todoStallPause and maxTodoStallRounds
• the todo_stall pause class and its RunPauseInfo mapping
• the planner's non-finalization message for it
• the Goal FSM's stop cause branch

trackTodoProgress no longer returns an error at all - it advances the streak and nudges once.

Verification

• TestTodoProgressGuardNeverPausesARun runs the identical stall under both ordinary chat and a Goal delivery scope, and asserts no run ends with a todo_stall pause in either. The chat case additionally asserts every scripted turn runs past the old threshold and that the reassessment nudge survived.
• TestTodoProgressGuardRenewsOnUniqueHostWork unchanged - the renewal rule is untouched.
• TestPauseClassNamesEachGuard updated: todo_stall is no longer a class.

go test -count=1 ./internal/agent/ ./internal/control/ ./internal/boot/ green; golangci-lint run ./... 0 issues; repolint clean with no baseline change (net removal).

Where this leaves the host

After #8320, #8340 and this, an ordinary chat turn has no host-owned stop of any kind. What remains is advisory (the zero-evidence ladder, the storm breaker, the exploration decay), user-invoked (--max-steps, the opt-in spend gate), or Goal's own bounded-autonomy contract.

Cache-impact: none - removes a host-side stop and its bookkeeping. No prompt, tool schema, or provider request field changes; the reassessment nudge is unchanged in wording and trigger.
Cache-guard: existing - go test ./internal/agent/ covers the cache-hit e2e suites, green unchanged.
Documentation-impact: none - the pause was never documented as a user-facing limit; docs/GUIDE.md describes the guards that remain, which are unchanged.