Let manual schedule fire honor runner wake budgets

[mostlydev]

Problem

claw api schedule fire is a synchronous, final-result-bearing operator command, but both client transport layers time out well before runner-specific wake budgets:

• cmd/claw-api/main.go local client mode defaults -request-timeout to 10 seconds.
• cmd/claw/api.go bounds the outer docker compose exec transport at 15 seconds.
• claw-api's wake adapter currently allows OpenClaw up to two minutes, and hermes: extend scheduled wake timeout #346 proposes the same evidence-based budget for Hermes.

A healthy manual wake can therefore be canceled by the client around 10 seconds even though the server-side adapter is allowed to keep waiting. The operator sees a timeout, the request context cancels the in-container exec, and schedule state can contradict the runner outcome.

Required behavior

1. Manual fire remains synchronous and returns the final scheduler outcome.
2. The local claw-api request timeout and outer compose-exec timeout cover the selected scheduler wake budget plus a small transport margin.
3. Read-only and fast schedule control commands keep short defaults; do not make every claw-api request wait minutes.
4. Timeout ownership stays in the operator/client and claw-api transport surfaces. Do not add runner prompt behavior, downstream wrappers, or cllama logic.
5. Cancellation before a wake attempt must remain neutral in schedule failure/degraded accounting (see Prevent slow scheduled wakes from blocking unrelated invocations #347).

Suggested shape

Give the fire subcommand an explicit longer transport budget and pass a matching -request-timeout to claw-api local client mode. Keep the generic --exec-timeout override available and document its precedence. If adapter-specific discovery is not available at the CLI layer, use a schedule-fire budget that safely covers the longest server-supported wake timeout.

Verification

• focused CLI argument/timeout tests for schedule fire
• other schedule operations retain their current short timeout
• an in-process delayed fire response beyond 15 seconds succeeds under the fire-specific budget
• go test ./cmd/claw-api ./cmd/claw
• go test ./...
• go vet ./...
• git diff --check

Relationship

PR #346 fixes Hermes server-side timeout selection but cannot fully resolve its manual-fire scenario until this client deadline is aligned. #347 separately handles scheduler concurrency and neutral cancellation.