perf(coro): amortize scheduler hot-path ownership

[cpunion]

Summary

• keep source-owner hot gates constant time
• batch up to eight adjacent timer or poll catalog entries per bounded reduction
• resolve exact same-P channel completions through an owner-local scheduler FIFO
• retain the managed-execution P lease across one bounded run slice and wake exact quota waiters
• release an inherited lease before physical-owner succession

The fast paths preserve the existing external A/ack/B source protocol for cross-P, callback, timer, poller, and foreign-thread producers. Same-P channel publication never resumes a peer recursively or performs typed cleanup while the channel lock is held.

Validation

• runtime internal/coro race and shuffled tests
• exact native fleet, program, hchan, JS/WASM hchan and JS/WASM program adapters
• complete runtime package tests
• compiler effect tests with and without llgo tags
• LLVM 22 native coroutine codegen and native fleet E2E
• standard-library time, file, pipe, TCP and DNS acceptance
• real JS/WASM time, freestanding file, host network, and TCP/DNS acceptance
• focused generated-binary retirement stress loops
• go vet for affected compiler and runtime packages

Performance, binary-size, and high-concurrency memory evaluation is intentionally deferred until this correctness change is integrated.