refactor(lsm): finish SSTable and table shard naming

Applies the SSTable / table shard / freeze vocabulary from the previous
commit to code that came in from main, and to spots that commit missed:

- `get_lsm_stats` test fixtures in Rust and Python still sent the old
  `buckets` / `generations` keys, so the Rust test target no longer
  compiled and the Python tests would fail against `table_shards`.
- The Java javadoc example called a nonexistent `LsmWriteSpec.tableShard`,
  an over-rename of `LsmWriteSpec.bucket`; the README used the old
  `buckets()` / `generations()` accessors.
- Prose that still said bucket (for a shard), seal, L0, or flushed
  generations, and identifier-style "table_shard" / "tableShard" in prose.

`LsmWriteSpec::bucket`, `num_buckets`, the hash-bucket values in the
merge-insert router, and the `"Sealed"` status wire value are unchanged:
those name the sharding transform or a server value, not the shard.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

nodejs/__test__/table.test.ts

@@ -4908,7 +4908,7 @@ describe("LSM convergence and stats", () => {
     const conn = await connect(tmpDir.name);
     const table = await lsmTable(conn);
 
-    // checkpointLsm seals first, so it surfaces flushLsm's rejection.
+    // checkpointLsm freezes first, so it surfaces flushLsm's rejection.
     await expect(table.checkpointLsm()).rejects.toThrow(/not supported/i);
   });
 });