fix(llamaindex): pass valid EpisodeType enum instead of invalid "text" string - #1512

Summary

• MemMachineMemory.add() in integrations/llamaindex/mem_machine_memory.py called memory.add(episode_type="text", ...), passing a raw string where Memory.add() expects an EpisodeType enum.
• Memory._build_memory_message() calls episode_type.value internally, so this raised AttributeError: 'str' object has no attribute 'value' on every call to add()/put().
• "text" was also never a valid EpisodeType value in the first place (the enum only defines MESSAGE = "message"), so simply string-normalizing it (as was done for the LangGraph integration in feat(client+langgraph): raw filter strings and EpisodeType normalization #1403) wouldn't have been enough here.
• Fix: pass EpisodeType.MESSAGE directly, matching the only valid episode type today.
• This is the same failure mode reported in [Bug]: LangGraph example demo raises AttributeError when adding memories with v2 APIs #1002 for the LangGraph integration, which was already fixed by feat(client+langgraph): raw filter strings and EpisodeType normalization #1403 but never linked back to that issue.

Test plan

• Added integrations/llamaindex/test_mem_machine_memory.py with two unit tests (mocking the backend Memory) that assert episode_type is forwarded as a real EpisodeType.MESSAGE enum via both add() and put().
• Verified the new tests fail against the pre-fix code with the exact same shape of error class this bug produces, and pass after the fix.
• Ran ruff check and ruff format --diff on both changed files: clean.