fix: add structured metadata filtering to CZI dataset search

[LiudengZhang]

What

Fix search_czi_datasets to actually use its organism and tissue filter parameters, and warn when filters are relaxed instead of silently dropping them.

Why

When the agent calls search_czi_datasets({"query": "lung, Mus musculus", "n_datasets": 5}), the top results are irrelevant (embryo and skin datasets) because:

1. The docstring says "input is a string containing: tissue, condition, and organism", which tells the agent to pack everything into query and never pass organism/tissue separately.
2. Even when filters are passed, they are silently skipped if the filtered set has fewer than n_datasets rows, with no warning.

How

1. Updated the docstring and Field descriptions to instruct the agent to pass organism and tissue as dedicated parameters, keeping query for condition/context only.
2. Replaced the silent filter drop with controlled relaxation: try organism+tissue first, fall back to organism-only, then to unfiltered. Each relaxation step includes a warning in the output so the agent knows the results are broader than requested.
3. Updated the embedding cache key to reflect actually-applied filters (avoids stale cache hits when a filter was relaxed).

Closes #5