fix: ComprehendFilterAgent client initialization and on_llm_start docstring

[abhu85]

Summary

This PR fixes two bugs reported in issues #359 and #360:

Issue #359 - ComprehendFilterAgent missing client initialization

When creating a ComprehendFilterAgent without passing an explicit client (relying on boto3 to create one), the agent would fail with AttributeError: 'ComprehendFilterAgent' object has no attribute 'comprehend_client'.

Root cause: Lines 34 and 39 in comprehend_filter_agent.py incorrectly assigned the boto3 client to self.client instead of self.comprehend_client.

Fix: Changed self.client to self.comprehend_client in both branches of the initialization logic.

Issue #360 - Documentation bug in on_llm_start callback

The docstring for on_llm_start in agent.py documented parameters that didn't match the actual method signature:

• agent_name (should be name)
• messages (not a parameter of this method)

Fix: Updated docstring to accurately reflect the method signature and context (LLM vs agent).

Test plan

• Added regression test test_initialization_without_client_uses_comprehend_client to verify comprehend_client is properly set when initializing without explicit client
• All 11 tests in test_comprehend_agent.py pass
• Verified docstring now matches method signature

Reproduction (Issue #359)

Before this fix, the following code would fail:

filter_agent = ComprehendFilterAgent(
    ComprehendFilterAgentOptions(
        name="Filter agent",
        description="Analyzes content",
        client=None,
        region="us-east-1"
    )
)
await filter_agent.process_request("test", "user", "session", [])
# AttributeError: 'ComprehendFilterAgent' object has no attribute 'comprehend_client'

After this fix, the agent correctly initializes and processes requests.

Fixes #359
Fixes #360