What's the problem?

search_memory iterates over memory entries from the Vertex AI API but had no error handling. If one entry was malformed or the iterator threw mid-stream, the whole call would blow up and drop everything collected so far.

What changed?

Added two layers of defense inside the iterator loop:

• Per-entry (AttributeError): skips bad entries and logs a warning so the rest still come through
• Iterator-level (Exception): if the stream fails partway, we return whatever we already collected instead of raising

Also added guards for a few edge cases I noticed while looking at this:

• memory is None
• fact is empty/None
• update_time is None (just sets timestamp=None rather than crashing)

Test plan

• test_search_memory — happy path unchanged
• test_search_memory_empty_results — empty results unchanged
• test_search_memory_skips_entry_with_none_memory
• test_search_memory_skips_entry_with_empty_fact
• test_search_memory_handles_missing_update_time
• test_search_memory_skips_malformed_entry
• test_search_memory_returns_partial_results_on_iterator_error