Per-schedule `timeout_seconds` override

`agent.toml` has a single `timeout_seconds` at the `[agent]` level. But many agents have multiple schedules with very different durations — `morning` might run a quick health check (10s) and `nightly` might run a full backup (30 min). Forcing both into the same timeout means either the quick one waits too long to fail or the slow one gets killed unfairly.

## Proposal

Allow `timeout_seconds` to be set per `[[schedules]]` entry, overriding the agent-level default:

```toml
[agent]
timeout_seconds = 60 # default for all schedules

[[schedules]]
id = "morning"
type = "cron"
weekday = "mon-fri"
at = "10:15"
# inherits 60s

[[schedules]]
id = "nightly"
type = "cron"
weekday = "mon-sun"
at = "02:00"
timeout_seconds = 1800 # override: 30 min for the heavy backup
```

## Acceptance criteria

- [ ] Schedule-level `timeout_seconds` parses and overrides the agent-level value at runtime.
- [ ] When absent, behavior is identical to today.
- [ ] `dotagent doctor` reports both values in `inspect`.
- [ ] [`agent-spec.md`](docs/reference/agent-spec.md) updated.

## Where to start

- `crates/dotagent-core/src/manifest.rs` — `Schedule` struct.
- `crates/dotagent-runner/src/lib.rs::run` — timeout resolution.

## Non-goals

- Dynamic timeouts (e.g. "10× the rolling average duration"). Static config only.

Start with the Schedule struct in crates/dotagent-core/src/manifest.rs, then follow timeout resolution from crates/dotagent-runner/src/lib.rs::run. Check the existing doctor inspect path and agent-spec.md. Done means schedule-level values override the agent default, omitted values preserve current behavior, inspect reports both values, and the reference documentation is updated.

Written by the indexing model from the issue text.