docs: v0.139.0 -- citation backfill batch 19: SRV (issue #164)

[SoundMatt]

Summary

Phase 2 (issue #164) citation-backfill batch 19: SRV (server.c's request-storage/admission/priority-scheduling/completion core).

Cited all 15 remaining uncited REQ-SRV-* requirements — SRV is now 100% cited:

• REQ-SRV-001-003 — pre-load-then-drain-on-enable queue (Table 32's ep_enable)
• REQ-SRV-004-005 — multi-request admission/rejection (§12.9.1.1 + Table 5)
• REQ-SRV-006-008 — priority-selection (§12.9.2's seven-tier ordering + safety-tagged persistence, cross-referencing the already-cited REQ-E2E-012/013 gate + FIFO tiebreak)
• REQ-SRV-009-011 — completion + trigger-occurrence recording (§12.9.3 Table 26)
• REQ-SRV-012 — chain-predecessor marking (§11.2.2.4)
• REQ-SRV-013 — cancellation (reusing Phase 1's established basis)
• REQ-SRV-014 — watchdog purge (the 0x0F/0x8F watch-dog-overflow rule already used for CMP)
• REQ-SRV-021 — pending-count getter (§12.9.2's "EP request storage" framing)

Also records a scoping finding: PWR (powerstate.c, 15 uncited) was investigated first and rejected as a batch candidate — a thin client-side wrapper over already-cited primitives (ep_wakeup.h + power.h), no distinct TC18 text of its own (same class as MDIO in batch 18).

One citation-text error caught and fixed before any edit was written: rcp_e2e_request_may_execute() was initially attributed to REQ-E2E-014 by proximity guess; grep against src/e2e.c's real tags showed it's REQ-E2E-012/013.

Verification

• Purely additive .fusa-reqs.json change; no source/test code touched.
• Full build: 69/69 targets. Full test suite: 64/64 passed.
• cfusa check: 0 errors.
• cfusa trace --req-coverage 100 --sec-tested 100: 100%/100%.
• .fusa-reqs.json JSON-validated.
• Citation count: 583/1028 (56.7%) → 598/1028 (58.2%). SRV module: 7/22 → 22/22 (100%).

Ref: issue #164.