worker: expose --heartbeat through hq alloc add

[MauriceDHanisch]

Summary

• hq worker start already accepted --heartbeat, but hq alloc add slurm/hq alloc add pbs had no way to set it — the field lived on WorkerStartOpts, not the SharedWorkerStartOpts that autoalloc's generated worker command reuses.
• Moves heartbeat onto SharedWorkerStartOpts (Option<Duration>, no default so we can detect "not set") and forwards it into the generated hq worker start ... invocation in autoalloc.rs only when explicitly provided, preserving the existing 8s default otherwise.

Motivation

On a busy server (e.g. mid MILP scheduler solve, or handling a burst of other workers registering), the single-threaded executor can go long enough without polling that an already-connected worker's heartbeat looks lost even though nothing is actually wrong. Autoalloc-managed workers had no way to raise their heartbeat interval to ride this out, since the flag was never wired through hq alloc add. Plain hq worker start was unaffected — this only closes the gap for the autoalloc path.

Test plan

• cargo check -p hyperqueue -p tako
• cargo test -p tako (no regressions)
• Manually verified hq alloc add slurm --heartbeat 45s renders --heartbeat "45s" in the generated sbatch worker command