feat(strands-memory): add converter injection and optional restored-tool filtering

[dharamendrak]

Issue #, if available:
#254

Description of changes:

• Added injectable message converter support in AgentCoreMemorySessionManager to remove hardcoded converter dependency and enable custom STM canonical formats.
• Introduced MemoryConverter protocol with:
  • message_to_payload(...)
  • events_to_messages(...)
  • exceeds_conversational_limit(...)
• Updated session manager constructor to accept optional converter and default to AgentCoreMemoryConverter for backward compatibility.
• Replaced hardcoded converter calls in write/read paths with self.converter.
• Added converter modules under integrations/strands/converters/:
  • openai.py (OpenAI canonical format converter)
  • protocol.py (shared interface and size-limit utility)
• Added optional config flag filter_restored_tool_context (default False) to strip historical toolUse/toolResult content blocks during restore when enabled.
• Streamlined converter initialization — default AgentCoreMemoryConverter is now set once in init instead of repeated self.converter or AgentCoreMemoryConverter fallbacks in each method.
• Added tests for:
  • OpenAI converter behavior
  • Session manager converter injection and restore filtering
• Updated comments/docstrings to clarify wording as "Strands-native message shape".

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

Testing

Validated the AgentCoreMemorySessionManager with and without OpenAIConverseConverter use below snippet

from bedrock_agentcore.memory.integrations.strands.converters import OpenAIConverseConverter
from bedrock_agentcore.memory.integrations.strands.session_manager import AgentCoreMemorySessionManager

  session_manager = AgentCoreMemorySessionManager(
      agentcore_memory_config=config,
      converter=OpenAIConverseConverter,
      region_name="us-west-2",
  )

Listed event from agent core memory: Raw AgentCore event confirming TOOL role is correctly persisted:

  {
    "memoryId": "strands_agentcore_memory_20260228_155315-R7WGNA4qCH",
    "actorId": "math_specialist",
    "sessionId": "math_20260304_181716",
    "eventId": "0000001772677043559#7ab31b1a",
    "eventTimestamp": "2026-03-04 18:17:23.559000-08:00",
    "payload": [
      {
        "conversational": {
          "content": {
            "text": "{\"role\": \"tool\", \"tool_call_id\": \"call_cRscVeeLyRbImBK4aQZcyCEG\", \"content\": \"Result: 425\", \"status\": \"success\"}"
          },
          "role": "TOOL"
        }
      }
    ],
    "branch": {
      "name": "main"
    }
  }

[afarntrog]

Have you had a chance to test this against some of the data to ensure it's working e2e?