use uuid v7 for time-ordered server-generated ids

[fabracht]

Closes #28.

Summary

• server-generated ids now use a UUID v7 base (48-bit millisecond timestamp + random) so ids are lexicographically time-ordered and prefix scans over data/{entity}/ return records in insertion order; the partition-targeting suffix loop is unchanged
• generate_id_for_partition drops its now-obsolete node_id/data parameters (UUID v7 randomness supplies uniqueness) — a breaking mqdb-core API change; all call sites updated and the two trivial cluster wrapper methods removed
• existing hash-based ids stay valid; they just don't sort chronologically alongside new ids
• wasm builds keep compiling — the v7 feature was added to the wasm uuid override, and the function is never called on wasm

Note: #90 (narrowed stale-index scan) was dropped from this PR — a quorum investigation showed it gives no perf benefit in a healthy database (the per-field subtrees union to the same key set as the entity-wide prefix) and narrows self-heal coverage when an entity's index definition is replaced. The entity-wide scan is kept.

Test plan

• core: generated id base parses as uuid v7, ids unique across calls, partition helpers unchanged
• core + agent + cluster suites green, clippy clean on all targets, wasm compiles