[fix][meta] Fix RocksdbMetadataStore instanceId not advancing across restarts

[SongOf]

Motivation

RocksdbMetadataStore#loadSequentialIdGenerator is supposed to persist
the initial value of the sequential-id counter to SEQUENTIAL_ID_KEY
when no such record exists yet in RocksDB. Instead, it wrote that value
to INSTANCE_ID_KEY by mistake.

Since the constructor calls loadSequentialIdGenerator() before
loadInstanceId(), this bug clobbers INSTANCE_ID_KEY to 0 right
before loadInstanceId() reads it back — so instanceId is always
computed as 0 + 1 = 1 on startup, instead of advancing (0, 1,
2, ...) across restarts, whenever no sequential node has ever been
created in that store.

instanceId is used to tag ownership of ephemeral nodes
(metaValue.owner == instanceId), so a store stuck on the same
instanceId across restarts loses the ability to distinguish the
current process instance from a previous one for ephemeral-node
ownership checks.

Modifications

• In RocksdbMetadataStore#loadSequentialIdGenerator, write the
  initial counter value to SEQUENTIAL_ID_KEY instead of
  INSTANCE_ID_KEY.

Verifying this change

• Make sure that the change passes the CI checks.

This change is already covered by existing tests, such as
RocksdbMetadataStoreTest#testMultipleInstances and
RocksdbMetadataStoreTest#testOpenDbWithConfigFile, which exercise
store open/close/reopen cycles.

Does this pull request potentially affect one of the following parts:

• Dependencies (add or upgrade a dependency)
• The public API
• The schema
• The default values of configurations
• The threading model
• The binary protocol
• The REST endpoints
• The admin CLI options
• The metrics
• Anything that affects deployment