[fix][metadata] Record get op stats on the correct completion branch in AbstractMetadataStore

[SongOf]

Motivation

Since #24580, the whenComplete callback in AbstractMetadataStore#get has its
branch contents inverted. The failure branch (t != null) calls
v.ifPresent(...), but on exceptional completion v is always null, so the
callback throws a NullPointerException that short-circuits
metadataStoreStats.recordGetOpsFailed. Per whenComplete semantics the NPE is
only attached to the original exception as a suppressed exception, so callers
still see the original failure — which is why this went unnoticed. The actual
damage is to observability:

• Failed get operations are never recorded in the
  pulsar_metadata_store_ops_latency{type="get", status="fail"} histogram, so
  the metric silently reports zero failures even when the metadata store is
  unhealthy (the only path still counted is the synchronous invalid-path check).
• nodeSizeStats.recordGetRes — which [fix][broker]Fix never recovered metadata store bad version issue if received a large response from ZK #24580 intended to run on the success
  path (cf. recordGetChildrenRes in getChildren) — is never invoked at all,
  so node-size stats for get responses were never collected.

Modifications

• In AbstractMetadataStore#get, move recordGetOpsFailed back to being the
  sole action of the failure branch, and move
  v.ifPresent(... nodeSizeStats.recordGetRes ...) into the success branch
  where it was intended to run.
• Add AbstractMetadataStoreGetStatsTest covering both halves of the bug
  against a minimal AbstractMetadataStore test double.

Verifying this change

• Make sure that the change passes the CI checks.

This change added tests and can be verified as follows:

• Added AbstractMetadataStoreGetStatsTest#testGetFailureRecordsFailedOpAndPropagatesOriginalException:
  injects a failing storeGet and asserts the original exception propagates
  without a suppressed NPE and that the status="fail" histogram count
  increments — fails before this fix.
• Added AbstractMetadataStoreGetStatsTest#testGetSuccessRecordsNodeSizeStats:
  asserts MetadataNodeSizeStats#recordGetRes is invoked with the returned
  GetResult and the status="success" count increments — fails before this
  fix (recordGetRes was never called).

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

If the box was checked, please highlight the changes

• 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

The existing pulsar_metadata_store_ops_latency{type="get", status="fail"}
histogram starts reporting real values (it previously never recorded async get
failures), and node-size stats for get responses are now actually collected. No
metric names, labels, or types change.