fix(sessions): handle list content in AdvancedSQLiteSession browsing helpers

[anxkhn]

AdvancedSQLiteSession.get_conversation_turns() and find_turns_by_content()
assumed a stored user-message content is always a str. That holds for
plain-text turns, but a valid user message can also carry a list of structured
parts (for example multimodal input_text/input_image items), and the session
stores each item verbatim via json.dumps, so json.loads(...).get("content")
hands these helpers a list.

That leads to two problems in get_conversation_turns(), whose docstring promises
'content': User message content (truncated):

• When the list has more than 100 parts, content[:100] + "..." evaluates
  list + str and raises TypeError: can only concatenate list (not "str") to list. The surrounding except (json.JSONDecodeError, AttributeError) does not
  catch TypeError, so the exception propagates out of asyncio.to_thread and
  aborts the whole call.
• When the list has 100 or fewer parts, the len(content) > 100 check is False,
  so the raw list is returned in the content field, which contradicts the
  documented truncated-preview string.

find_turns_by_content() shares the same string-only assumption and returns the
raw list in content too.

The fix adds a small module-level _content_preview helper that coerces non-string
content to a JSON string (json.dumps(content, ensure_ascii=False)) before applying
the existing truncation. get_conversation_turns() uses it with a 100-character
limit and find_turns_by_content() uses it without a limit, so each keeps its prior
behavior for string content exactly while now returning the documented preview
string for list content. full_content still keeps the original value in both
helpers. This mirrors the defensive handling the sibling _validate_turn already
does for its own preview slice.

Test plan

Added two regression tests in
tests/extensions/memory/test_advanced_sqlite_session.py (the existing suite only
exercised plain-string content):

• test_get_conversation_turns_with_list_content adds a short list-content turn and
  a 101-part list-content turn, then asserts get_conversation_turns() returns
  without raising, that content is a str for both, that the long one is
  truncated with a trailing ..., and that full_content keeps the list.
• test_find_turns_by_content_with_list_content asserts a list-content match returns
  a str preview in content while full_content keeps the list.

Verified red -> green: reverting only the source change makes
test_get_conversation_turns_with_list_content fail with the TypeError and
test_find_turns_by_content_with_list_content fail because content is a list;
with the fix both pass. Also confirmed the string path is unchanged with an
equivalence check across empty, short, length-100, length-101, longer, and unicode
inputs.

Local checks:

• uv run pytest tests/extensions/memory/test_advanced_sqlite_session.py -> 45
  passed.
• uv run pytest tests/extensions/memory/ -> 216 passed, 5 skipped
  (external-service integrations).
• make format (no changes), make lint clean, pyright on the changed module: 0
  errors.

Issue number

N/A

Checks

• I've added new tests, if relevant
• I've run .agents/skills/code-change-verification/scripts/run.sh
• I've confirmed all verification steps pass
• If using Codex, I've run /review before submitting this PR