Reconcile module naming with RELAY spec v1.14 §13.7.2

[SoundMatt]

Addresses #87.

Summary

Reconciles this repo's TC18-replacement module naming with the standard
module-name registry RELAY spec v1.14 added to §13.7.2, per the deltas
identified in #87 (independently re-verified against the current code
before acting on each one):

• avtp.h, acf.h, discovery.h, regmap.h — already correctly
  named. Confirmed, no change.
• safept.h/safept.c merged into e2e.h/e2e.c, replacing the old
  ad-hoc CRC-16 + sequence-counter + replay-guard wrapper outright, matching
  the equivalent merges already shipped in cpp-RCP's e2e.hpp and
  rust-RCP's e2e.rs (both confirmed via their public source: full content
  replacement, no legacy shim, since nothing else in either tree depended
  on the old CRC-16 API — true here too). Renamed rcp_safept_*/
  RCP_SAFEPT_* to rcp_e2e_*/RCP_E2E_*. Renumbered the surviving 27
  requirements in .fusa-reqs.json as REQ-E2E-001..027, mirroring
  cpp-RCP's own precedent of renumbering a merged module's requirement ids
  under the new module name; removed the 12 old CRC-16-era REQ-E2E-*
  entries whose implementation no longer exists.
• server.h/server.c split: the RC Server lifecycle state machine
  (registry's lifecycle concern) moved to new lifecycle.h/lifecycle.c
  with an rcp_lifecycle_* prefix, matching cpp-RCP's own
  REQ-LIFECYCLE-* requirement-id precedent for the equivalent module. The
  per-endpoint ep_enable pre-load-then-drain queue mechanism — a
  distinct, registry-unconstrained concern neither cpp-RCP nor rust-RCP
  have an equivalent module for yet — stayed under the original server.h
  name. test_server.c split into test_lifecycle.c + a reduced
  test_server.c the same way. REQ-SRV-001..020,024 → REQ-LIFECYCLE- 001..021; REQ-SRV-021..023 (queue-only) → REQ-SRV-001..003.
• Conditional-request taxonomy grouped under a common request_
  prefix: request_cancel.h, request_chained.h, request_compound.h,
  request_timed.h, request_triggered.h, request_sequencer.h — the
  registry's request entry names "compound, compound-wait, triggered,
  chained, timed... and sequencers" explicitly. This matches the codebase's
  own existing ep_*.h grouping idiom for "one registry name, several
  per-language files." scheduler.h stays unrenamed: it's a distinct
  execution-priority concern the registry doesn't name (and which
  safept.h's own file header already called out as "a separate,
  unduplicated concern" from the request taxonomy) — the original issue's
  bullet lumped it in with the other six, but the registry text itself
  doesn't. Internal function/type prefixes (rcp_cancel_*,
  rcp_compound_*, ...) are unchanged; only file grouping changed, per the
  issue's own "packaging is idiomatic per language" framing.
• fragment: no dedicated module exists yet — Phase 20
  fragmentation/reassembly is unimplemented (acf.h's
  read_size_or_segment_num is round-tripped only). Nothing to rename;
  noted in e2e.h's file header for when Phase 20 lands it.
• canbr/linbr ADAPT-vs-DEPRECATE: left untouched, exactly as the
  issue flagged. For the record: this repo's own ROADMAP.md already
  independently commits to ADAPT (narrowed role) for both, matching
  go-RCP and diverging from cpp-RCP's DEPRECATE — that's a pre-existing
  decision this PR doesn't touch or resolve.

Scope

Renames, a merge, and a split only — no protocol behavior changed.

Verification

• Full build: 0 warnings under -Wall -Wextra -Wpedantic (both the
  library and the -DRELAY_BUILD_CLI=ON CLI target)
• Full test suite: 61/61 passing
• relay conform --strict against the built CLI: PASS
• cfusa check/lint/analyze/cyber/trace --req-coverage 100/verify/qualify/ vuln: all pass, except cfusa check's pre-existing HARA002/HARA003
  findings against .fusa-hara.json — confirmed identical (same 20
  errors) on an unmodified main checkout, so unrelated to this PR.
• .fusa-reqs.json requirement catalog stays in exact bidirectional sync
  with every //cfusa:req///cfusa:test tag in the tree (804/804 both
  directions).

Known, out-of-scope gap

tara.md/CYBERSECURITY.md/SAFETY_PLAN.md/FORMAL_VERIFICATION.md
still reference the old REQ-E2E-004..012 ids (replay-guard/CRC threat
mitigations) that no longer exist post-merge. cpp-RCP's shipped docs have
the identical latent gap after its own equivalent merge. Updating
threat-model documentation to reflect the CRC32-only mechanism is a
security-review judgment call, not a naming fix, so it's left for a
dedicated follow-up rather than folded into this PR — noted directly in
e2e.h's file header for visibility.