Add two Topic Reader state gauges

[KirillKurdyukov]

Summary

Add two Topic Reader observable gauges, using the closeable API merged in #735:

• ydb.topic.reader.partition_session.count, unit {session}.
• ydb.topic.reader.credit_balance_bytes, unit By.

After successful stream initialization, ReaderMetrics.registerStream(stream)
registers callbacks reading that concrete stream. Each replacement closes both
previous registrations. Retry/end unregister the matching stream; Reader close
removes registrations permanently.
Registration takes a nonnull stream. A private closeGauges() performs shared cleanup;
there is no registerStream(null) command. Final close() also rejects future registrations.

No changes to TopicRetryableStream and no metric getters in ReaderImpl.
ReaderMetrics owns two registration fields, without a composite lambda or handles
on ReadSession. No additional accounting maps/counters or reading/retry changes.

Names, units and descriptions match the C# SDK. The two existing received-counter
descriptions are also aligned. Common attributes remain consumer and reader.name.

No Core, table/query pool or public settings changes.

Validation

Java21 clean package passed46 tests, no failures/errors/skips:

JAVA_HOME=$(/usr/libexec/java_home -v 21) mvn -B -pl topic -am \
  '-Dtest=OpenTelemetryMeterTest,ReaderMetricsTest,ReaderImplTest,SyncReaderImplTest,AsyncReaderImplTest,BufferManagerTest' \
  -Dsurefire.failIfNoSpecifiedTests=false clean package

All Reader metric tests are in ReaderMetricsTest: the existing counter case,
one gauge happy-path case, and a stale-source/closed-owner regression, with a shared Meter.
After extracting cleanup, all three ReaderMetricsTest cases and the SDK/real OTel
smoke were rerun successfully, including late init and concurrent shutdown.

Public TopicClient/SyncReader smoke with a real OpenTelemetry collector:

• No gauge observations until successful stream initialization.
• Partition count1; credit100->80->100 on receive/delivery.
• Repeated init does not add callbacks.
• Retry removes the old observations; reconnect registers the initialized replacement.
• Closing one reader leaves the other; closing both leaves no gauge observations.
• No sleeps/polling: the captured scheduler retry is invoked directly.
• Shutdown from the init completion callback and a later InitResponse do not recreate gauges.
• Native credit registration paused by an event while shutdown closes the RPC stream:
  after releasing/joining both workers, the real collector has no gauge observations.

LSP diagnostics and javac/checkstyle/package checks pass. Expected deprecated-API
warnings from compatibility tests remain.
Database-backed integration tests were not run locally: the Docker daemon is unreachable.
The SDK smoke uses fake network streams and the real native OpenTelemetry backend.

Lifecycle tradeoffs

• A metric-only synchronized operation serializes the two native registrations and
  their close calls; no lock is added to message processing, counters or Reader state.
  Lifecycle callers can wait for a concurrent registration/backend operation.
• One active stream per ReaderMetrics. Independent Readers use independent instances;
  a future federated reader must not share this single slot across concurrent streams.
• Identity-checked unregister prevents old stream completion removing a newer source.
  The existing ReadSession closed flag and a terminal metrics flag reject late init.
• Observations are absent before init/in retry gaps, rather than zero.
• Gauge replacement is not a transactional two-instrument collector snapshot.
  Actual unregister requires a backend implementing closeable registrations;
  legacy Meter's default NOOP registration remains the existing compatibility limitation.

Tracker: https://st.yandex-team.ru/YDBAPPTEAM-2094.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 75.07%. Comparing base (01a0b6b) to head (f235c2a).
⚠️ Report is 4 commits behind head on master.

Additional details and impacted files

@@             Coverage Diff              @@
##             master     #734      +/-   ##
============================================
+ Coverage     74.70%   75.07%   +0.36%     
- Complexity     3669     3689      +20     
============================================
  Files           394      394              
  Lines         16633    16639       +6     
  Branches       1757     1755       -2     
============================================
+ Hits          12426    12491      +65     
+ Misses         3593     3533      -60     
- Partials        614      615       +1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.