perf(coro): fuse logical task runtime state

[cpunion]

Summary

• fuse each spawned scheduler G and logical runtime context into one exact rooted task envelope
• keep physical M/P only in executor or native contexts and lend them to a logical G for the physical resume interval
• preserve the compiler spawn ABI with G at allocation offset zero and keep static command context ownership separate
• add architecture source gates and record the rejected BDWGC size-class variants and final benchmark

Validation

• full runtime test suite
• all 20 native fleet E2E tests
• linked defer, panic, channel and spawn E2E tests
• native standard-Go compute, spawn, park, handoff, timer and parallel workloads
• JS/WASM execution under Node
• WASI malloc and WASI tinygogc execution under Wasmtime
• JS/WASM, Linux ARM and RISC-V, and Cortex-M baremetal adapter builds

Performance

• native64 dynamic task envelope: 264 + 104 = 368 bytes, down from separate BDWGC classes totaling 432 bytes
• 10,000 parked G median RSS: 29,409,280 bytes versus Go 31,539,200 bytes
• incremental cost at 10,000: about 2,166 B/G versus Go 2,802 B/G
• exact-parent interleaved 10,000-G RSS: 30,375,936 to 29,442,048 bytes, down 3.07 percent
• final short-workload medians show compute and handoff neutral with no displaced regression; stripped artifact grows 992 bytes