Faster tuist cache warm with parallel XCFramework creation

XCFramework creation during tuist cache warm now runs concurrently across targets, with the fan-out bounded by the host’s active processor count and capped at 8. The “Creating XCFrameworks” phase is no longer a serial bottleneck. On a 150-target sample, the phase drops from around 111s to roughly 25s, and on a large cache warm of framework targets the bottleneck phase has been measured shrinking from roughly 9.8 minutes (about 564s) to under 3 minutes.