Make message filtering in conversation search workable in sqlite3

[lemorage]

Please describe the purpose of this pull request.
The agent's conversation_search tool, can search relevant content in the past conversations, by the given query_text. However, this can only work in PostgreSQL, since the ILIKE and json_array_elements are PostgreSQL-specific syntaxes.

letta/letta/services/message_manager.py

Lines 307 to 316 in 6733849

	if query_text:
	content_element = func.json_array_elements(MessageModel.content).alias("content_element")
	query = query.filter(
	exists(
	select(1)
	.select_from(content_element)
	.where(text("content_element->>'type' = 'text' AND content_element->>'text' ILIKE :query_text"))
	.params(query_text=f"%{query_text}%")
	)
	)

This PR adds the SQLite3-compatible syntax, to fix #2553.

NOTE: This PR uses dialect branching to resolve the problem; there might be a more elegant and unified way to handle this, but I did not think of any. Welcome to suggestions!

How to test
poetry run pytest -s tests/test_managers.py -k test_list_messages_with_query_text

Have you tested this PR?
I added a unit test to reproduce this error. If you ran letta server locally, the test would fail without this fix. Also, the reproduction steps mentioned in #2553 can be used for testing.

[mattzh72]

Thanks for the contribution!