Avoid full task-list scans for async identity lookup and dependent wakeups

[LunaStev]

The executor performs linked-list scans in _find, _wake_dependents, and each _drive scheduling round. Task records are retained for reuse until shutdown. After a large fan-out has completed, operations can still traverse the historical high-water list; waking many dependents also produces quadratic work in the number of task slots.

This is a source-level complexity finding, not a claimed benchmark result. A focused patch should establish deterministic work counters before choosing data structures.

Source evidence at the head of #520:

• std/task.wave:167 — fun _find
• std/task.wave:175 — fun _wake_dependents
• std/task.wave:306 — export(c, "__wave_task_drive")

Acceptance:

• Add bounded fan-out/join and grow-then-shrink workloads with operation or traversal counters rather than timing-only gates.
• Provide efficient identity lookup and explicit dependency/ready bookkeeping while keeping the single-threaded executor model.
• Ensure completed/reused slots do not make steady-state work proportional to all historical slots.
• Preserve single-consumer futures, wake coalescing and cancellation correctness.

Related: #444. This is a focused complexity follow-up to its initial executor, not a request for a multithreaded runtime.

Audit status: identified by static source inspection; the scenarios above have not been executed during this audit. This follows up the implementation introduced in draft PR #520.