fix(mpc): cap Energyplan time budget by remaining slot time #1148

#1130 refuses to publish a plan whose first price slot ended during the solve. The worker budget is still model-size only (500ms vs 5s), so a late-in-slot replan can burn the full budget, get slot_elapsed, and retry.

Found on master 6eadec3f in go/internal/mpc/energyplan.go and execution_time.go.

What happens

energyplanTimeBudget looks at slot×asset size and PV curtailment. Request TimeLimitS = min(Timeout*0.8, budget). Process timeout is 7s.

Remaining first-slot time is not used. trimFirstExecutionSlot runs before solve; after solve, firstSlotExpired discards the plan and queues slot_elapsed. A 5s fleet solve started with 200ms of slot left spends the budget on a result that cannot be published.

Replan ctx is WithCancel only — no wall deadline tied to the slot end.

#1144 is the rain-check LenMin leftover. This is the worker wall-clock leftover.

What a fix needs

• Cap TimeLimitS (and the process wait) by remaining first-slot time plus a small publish margin.
• If remaining time is below the small-model budget, skip or fail fast rather than solve and discard.
• Test: now = slot end − 200ms, fleet size that would get 5s — worker budget ≤ remaining, no slot_elapsed retry from a plan that could not have landed.

Open PR overlap: #1131 is changelog-only for #1130. #1129 is storage.