Set a build queue's disk floor from the measured incremental cost

A local build queue that skips jobs below a fixed free-disk floor kept skipping almost every job: each build's leftovers pushed free space just under the floor by the time the next job was checked. Measure what an incremental build actually consumes (here 1-2 GB in an already-warm DerivedData), and set the floor to the hard-failure point plus that cost plus a margin, not to a round 'comfortable' number. Keep the higher number as a reminder for humans, not as a hard gate. With a LIFO worker, a skipped job is also dropped from the queue, so a too-high floor silently loses work as well as delaying it.