Let hflow catalog ui explore bucket-backed catalogs

[kstonekuan]

Current behavior

hflow catalog ui rejects every bucket URL before startup at src/hflow/cli.py:1287-1300. CatalogUiSettings and the polling lifecycle also require a local Path at src/hflow/catalog_ui.py:25-39 and src/hflow/catalog_ui.py:93-152.

open_catalog_connection() already accepts bucket catalog roots and syncs their append-only Parquet tables into the local mirror at src/hflow/curation.py:221-233 and src/hflow/curation.py:260-281. The browser is the remaining local-only catalog surface.

Definition of done

• hflow catalog ui --catalog s3://..., gs://..., and az://... start against existing HFlow catalogs when the bucket extra and provider credentials are available.
• Startup validates the existing remote catalog marker. Browsing never creates or changes objects in the bucket.
• Catalog Parquet files sync into the existing local mirror before the DuckDB connection opens.
• The polling lifecycle syncs newly appended catalog files and makes the first and later completed appends visible in the open UI without a restart.
• A valid empty bucket catalog starts the UI and waits for its first completed append.
• The startup output names the bucket catalog clearly and doesn't present the mirror directory as the source catalog.
• DuckDB continues listening on loopback. SQL reads the local mirror and gets no direct object-store write surface.
• Missing bucket dependencies, credentials, and remote catalog markers produce concise CLI errors with no traceback.
• Add behavioral tests covering a remote catalog at startup, an empty remote catalog receiving its first append, a later append becoming visible, and no bucket writes during browsing.
• Update docs/CATALOG.md with bucket usage, mirror behavior, and the existing loopback/SSH guidance.

Use the catalog's current append and mirror semantics. This issue doesn't introduce a stronger snapshot-isolation contract.

Non-goals

• A custom HFlow frontend
• Listening on a public network interface
• Editing bucket objects through DuckDB SQL
• Catalog compaction or a new catalog format

Validation

uv run ruff check
uv run ruff format --check
uv run ty check
uv run pytest tests/test_catalog_ui.py tests/test_cli.py tests/test_storage.py -q
uv run pytest -q