Monte Carlo finish-date forecast

When at least one task has a PERT estimate (see "PERT — three-point estimates"), the Overview tab gains a **Monte Carlo forecast** card.

What it shows:
• **P50 / P80 / P95** — the project end date at three confidence levels. P50 is the median (50% chance of finishing by then); P80 is conservative; P95 is the "worst case I'd commit to" date.
• **Probability of finishing on time** — chance of finishing by the authored end date, computed across 1000 simulated trials.

How it works:
For each trial, every PERT-estimated task has its duration resampled from a triangular distribution (close approximation of the PERT beta distribution; much faster). The forward-pass CPM is re-run, and the project end is recorded. After 1000 trials, the distribution of end dates yields the percentiles.

Why three percentiles?
• Tell stakeholders the **P50** when you want a realistic single date
• Use **P80** for committed external dates
• Reserve **P95** for worst-case planning (insurance, contingency)

Skipped when no tasks have PERT — without uncertainty estimates the simulation has no signal.

Wally can run it on demand: "What's the probability the renovation finishes on time?" calls `project_simulate` and returns the answer.