MXD v0.2.7

The end-to-end smoke of v0.2.5 on a fresh Ubuntu 24.04 VM crashed on
the first block store with:

./db/c.cc:532: bool SaveError(char**, const rocksdb::Status&):
  Assertion `errptr != nullptr' failed.

v0.2.6 fixed the symptom at the distribution layer (pin build to
22.04 + bundle librocksdb 6.11). v0.2.7 fixes the underlying chain-
code bug so letsgo --from-source paths on any rocksdb 7.x/8.x
distro also work cleanly.

Fixed

• src/mxd_blockchain_db.c — 7 sites converted from
  rocksdb_put/get(..., NULL) to the standard
  char *err = NULL; rocksdb_put(..., &err); if (err) { log; free; }
  pattern. All are best-effort housekeeping ops around the
  current_height and latest_stored_height keys:
  • store_block_unconditional — latest_stored_height put
  • store_block_unconditional — gap-fill probe (get)
  • store_block_unconditional — gap-fill current_height put
  • reorg_to_candidate — post-rollback current_height put
  • mxd_block_exists_at_height — existence probe (get)
  • mxd_advance_height_pointer — probe (get)
  • mxd_advance_height_pointer — current_height put
    Behavior on success path is unchanged. Behavior on failure path
    goes from "silent error swallow" (rocksdb 6.x) or "process abort"
    (rocksdb 8.x) to "log WARN + continue".

Verified

• CI green on both gcc + clang (37/37 tests, sanitizers clean).
• Coordinated deploy on 6-node testnet, 8-minute soak: NRestarts=0,
  consensus tick normal, zero rocksdb assertion lines in journals.

Notes

• Mainnet validators still run the v0.2.0 binary (no symptom there
  — rocksdb 6.11 tolerates the NULL). Mainnet rollout of this fix
  will happen at the next coordinated v0.3.x release with a
  staggered/canary deploy pattern.
• Other rocksdb-heavy files (mxd_utxo.c, mxd_contracts_db.c,
  mxd_transaction.c) already used the correct &err pattern;
  no changes needed.

---