Unify concurrency request chain identity

[udnay]

Motivation

• Eliminate holder leaks and limit-bypass bugs by giving queued concurrency requests a durable chain identity so scanner resumption can continue the same limit-walk instead of inventing a synthetic id.
• Make the enqueue/scanner/ticket flows share a single resumeable walker model so immediate grants, queued requests, and future RequestTicket promotions behave consistently.
• Keep the change focused: only unify request payloads/identity and scanner resumption behavior, leaving tombstone/task-key-start-ms conflict handling to the planned follow-up.

Description

• Persist chain resume state in request intents: added task_id, held_queues, next_limit_index, and serialized limits to ConcurrencyAction::EnqueueTask and the FlatBuffer schema, plus codec helpers to encode/decode limit entries.
• Stop writing interim runnable RunAttempt on immediate per-limit grants; append_grant_edits now writes only the holder record and the final RunAttempt is written by the unified walker once the chain fully resolves.
• Reworked grant scanner (process_grants) to decode the extended EnqueueTask, validate payload, reserve slots across possibly multiple queues while resuming the chain, emit new request edits when the resumed walk pauses on a later queue, derive per-pass request-counter deltas from walker actions, and roll back all in-memory reservations from the pass on write failure.
• Caller-side changes: enqueue_limit_task_at_index forwards held-queues/next-index to concurrency helpers so queued intents carry the full resume payload; query code prefers decoding the request payload's task_id when available; added a regression test scanner_resumes_paused_chain_with_original_task_id_and_all_holders and adjusted existing codec/tests to the new payload shape.
• Edge cases intentionally out of scope: tombstone / task_key_start_ms conflicts (handled by separate task_key_start_ms work), RequestTicket future scheduling semantics are not fully expanded to carry full chain state, cancel/reimport cleanup for EnqueueTask-held queues requires separate rework, scanner resumption uses durable/default floating capacity rather than inline floating refresh, and queue/index mismatch validations / exhaustive malformed-payload policies are minimal in this change.

Testing

• Ran quick static checks: git diff --check and cargo fmt --check — both passed.
• Attempted cargo check --tests / cargo test but the build script invokes flatc (FlatBuffers compiler) and the environment lacks flatc, causing the build script to panic and tests to be blocked; see build error "failed to run flatc … No such file or directory".
• Added/updated regression tests and codec round-trip coverage (notably scanner_resumes_paused_chain_with_original_task_id_and_all_holders, codec roundtrips for the extended EnqueueTask), but their execution was blocked by the missing flatc in CI/local environment.
• Manual code-level review and local cargo fmt/linting performed; no formatting or obvious compile-time syntax issues remain beyond the build-tool dependency failure.

---

Codex Task

[udnay]

This change is part of the following stack:

• Unify concurrency request chain identity #333 ◀

_{Change managed by git-spice.}