feat(logging): attach structured JSON logging with sessionId to root logger

[luyuzhe111]

Description of changes:

Problem

When running RL rollouts on ACR, logs from user code (rl_app.py, reward.py) and third-party libraries (migration_bench, httpx) were impossible to correlate in CloudWatch. The upstream SDK's RequestContextFormatter only attaches to the bedrock_agentcore.app logger — every other logger in the process used the default text formatter with no sessionId, making it impossible to filter all logs for a single rollout session.

Solution

• Add CorrelatedFormatter that emits structured JSON with sessionId and requestId pulled from BedrockAgentCoreContext (the public API)
• Attach it to the root logger in AgentCoreRLApp.__init__ via configure_logging() — every logger in the process inherits it automatically
• Users don't need to change anything beyond logging.getLogger(__name__) to get full session correlation

Changes

• src/agentcore_rl_toolkit/logging.py — new module with CorrelatedFormatter and configure_logging()
• src/agentcore_rl_toolkit/app.py — call configure_logging() in __init__, switch bare logging.info() to named logger
• Examples — print() → logger.info() in reward functions for structured output
• Honors user-specified log level (e.g., logging.basicConfig(level=DEBUG)) via min() logic

Test plan

• Unit tests (tests/test_logging.py) — formatter output, context injection, idempotency
• Local e2e — started server, submitted request, verified JSON output with requestId (migration agent)
• ACR e2e — deployed to AgentCore, ran evaluate_async.py --limit 1, confirmed sessionId appears in CloudWatch for all loggers (reward, root, __main__, agentcore_rl_toolkit.app) (migration agent)

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.