Apply async-with pattern to ValkeyClient with retry-based disconnection

[jopemachine]

Objective

Refactor ValkeyClient usage to an async with pattern that internally tracks retry counts and decides whether to tear down the connection based on consecutive failures.

Details

• Wrap ValkeyClient operations in an async with context manager pattern
• Track retry_count internally; when the threshold is exceeded, disconnect the client rather than continuing to retry on a broken connection
• Migrate all ValkeyClient usage sites to the new async with pattern
• This has a wide blast radius — all 14 domain-specific Valkey clients and their callers need updating

Implementation Reference

AgentClientPool (src/ai/backend/manager/clients/agent/pool.py) uses a similar pattern with acquire(), _record_failure(), and _record_success() that can serve as a reference for the implementation approach.

Code References

• MonitoringValkeyClient: src/ai/backend/common/clients/valkey_client/client.py (~line 413)
• Domain-specific clients (14 total): src/ai/backend/common/clients/valkey_client/valkey_*/client.py
• Manager Valkey dependency: src/ai/backend/manager/dependencies/infrastructure/redis.py
• Agent Valkey dependency: src/ai/backend/agent/dependencies/infrastructure/redis.py
• AgentClientPool (reference only): src/ai/backend/manager/clients/agent/pool.py

Acceptance Criteria

• ValkeyClient operations are wrapped in an async with context manager
• Retry count is tracked internally and the connection is torn down when the threshold is exceeded
• All ValkeyClient usage sites are migrated to the new pattern
• Verified in both Standalone and Sentinel modes
• Tests covering the retry-based disconnection logic are included

JIRA Issue: BA-5578