Reduce the cost of forcing a fresh LazyCell #399

Constructing and forcing a fresh LazyCell with an immediately ready initializer took 314 ms for seven million cells. This measures the first shared force_pin on each cell. The peer starts with a stored future, while LazyCell::new starts with a factory that produces one.

Apple M4 Max, macOS arm64, rustc 1.101.0-nightly. Each cell gives rank, elapsed time relative to Asyncband and the whole-batch median over 10 interleaved runs. Lower is faster. Supplemental measurements

force_pin locks the initializer state and drives the stored future before publishing its value. The macOS allocation probe observed one 64-byte allocation per fresh Asyncband cell and zero for async-once-cell. A separate run with only the internal lock switched to parking_lot took 135 ms versus 339 ms for the original, with the peer at 67 ms.

First contribution

Compare LazyCell::new with LazyCell::from_future using the same immediately ready value and fixed round count. Measure forcing an already initialized cell separately. For a change to shared initialization, verify that a second caller resumes the same pinned future after the first caller cancels, and that a panicking initializer still poisons the cell.

To run the supplemental peer_checks command below, save and run peer-checks.rb from the linked comment in a disposable checkout of 6102089. The benchmark and library sources match 36ba6f2. Generator and instructions

cargo x bench --bench peer_checks -- extra/lazy-cell-first-force --rounds 7000000 --repeat 10 --raw

Performance tracking issue #396