fix(storage): a silent set-mate cannot hold the mirror sync; no node source reads a clock; five route-chain rows re-verified

[cryptskii]

Storage batch 2: one liveness bug fixed, the node's no-clock rule enforced by a test, and five route-chain rows re-verified against main.

Implemented and proven in this PR

A silent set-mate held the mirror sync (§6.53).

• The node's set client had no connect or request timeout, and mirror_sync holds a node-wide one-at-a-time lock.
• A set-mate that accepted a connection and never answered held that sync, and every later one on the node, open indefinitely.
• The fix:
  • set_client: connect bound 5 s, request bound 10 s. A fetch that times out fails as an unreachable set-mate does (502, lock released). It stores and orders nothing, so no protocol fact reads the clock.
  • MAX_SYNC_CYCLES 1024 → 128, so one sync stays inside the SDK's 30 s request bound. Every cycle is kept as it is fetched, so the next sync continues where this one stopped.
• Test: dsm_storage_node::bytecommit_chain::a_set_mate_that_never_answers_fails_the_sync_and_frees_it.
• Mutation control: timeouts removed → red ("sync 0 hung on a set-mate that never answers").
• In the SDK, route_seats::read_cell now asks every seat at once (values, cycle closes, mirror syncs), with the answers kept in route order. A silent seat costs one client timeout instead of one per seat.

No node source reads a clock: MR-STOR-0008, Partial → Met.

• New test dsm_storage_node::no_clock_reads parses every node source with syn:
  • it reads paths and use trees whole, and macro bodies token by token;
  • it refuses Instant, SystemTime, UNIX_EPOCH, chrono and tokio::time;
  • comments and strings never count.
• Its self-test covers grouped, renamed and glob imports, a macro body, and a comment and string that must not count. A first token-window draft missed use tokio::{sync::Mutex, time}; the self-test caught it.
• Mutation controls, each red and then restored:
  • a grouped-import tokio::time sleep in the spool;
  • Instant::now() in db/pg.rs.
• ci/no_clock_and_no_json.sh loses two node exemptions: the spool (which reads no clock) and a rate limiter that no longer exists. A clock read in the spool now fails the script too (checked, then restored).
• Adds syn and proc-macro2 as node dev-dependencies. Both are already in the lock at the same versions; Cargo.lock changes only the node's dependency list.

Re-verified as already satisfied on main (records only)

These rows were Missing on notes written before route chains existed. Each cited test passes at this tree.

Row	Now	Evidence
MR-STOR-0131 links	Met	route_chain::evaluate; links carry position and prior chain
MR-STOR-0133 later-link validity	Met	leader chain, higher position, seat's mirror of the leader
MR-STOR-0138 Preserved ≠ Final	Met	every consumer treats Preserved as not final
MR-STOR-0094 ByteCommit not counted	Met	mirrors must agree; Core checks chain link and root itself
MR-STOR-0134 empties	Partial	no-response recorded and never counted; the writer never produces a taken empty

• MR-STOR-0140 also cites the silent set-mate test.
• VERIFICATION_MATRIX gains rows for the liveness test and the clock test, each with its mutation control.

Verification

• dsm_storage_node bytecommit_chain 8/0 and no_clock_reads 2/0 (release, Postgres).
• dsm_sdk route_seats 9/0 (release).
• cargo fmt --all --check and clippy -p dsm_storage_node --all-targets -D warnings are clean. scripts/real_code_guard.py, ci/no_clock_and_no_json.sh and ci/conformance_evidence.py pass.
• Branch is 0 behind main at d19ea36.

[cryptskii]

Included in #1089, which merged this branch's commits (and the trace fix 9dc6f84 for #1087) and repinned once at the combined head. One PR, one merge, at the owner's direction.