fix: _parse_path fails to navigate a lone top-level bracket index - #236

What does this PR do?

Fixes RichExplorer._parse_path() in src/deepset_mcp/tokonomics/explorer.py: when a navigation path has exactly one component, it returned the raw part (parts[0]) instead of wrapping it in glom.Path. That's fine when the lone part is a string attribute name (glom accepts bare strings as specs), but breaks when it's an int bracket index — glom does not accept a raw int as a spec, so glom(obj, 0) raises a TypeError wrapped in GlomError, which _get_object_at_path reports as Object 'X' does not have a value at path '[0]' — even though index 0 clearly exists.

This affects every object-store tool that takes a path parameter (get_from_object_store, get_slice_from_object_store, grep_object_store, sed_object_store, yq_object_store): whenever an LLM tries to index the first/Nth item of a top-level list result — a very natural first navigation move — it gets a false "not found" error instead of the value.

Bonus fix, same root cause: an empty bracket pair (path="[]") previously raised an uncaught IndexError (not GlomError) directly out of _parse_path, since _get_object_at_path's try/except GlomError only wraps the glom() call itself, not the _parse_path() call that builds the spec. Fixed by the same change.

Fix

Always use Path(*parts). glom.Path already handles a single part correctly (string or int) and zero parts (returns the object unchanged) — the special-casing was unnecessary and the source of both bugs.

How was this tested?

• Reproduced both bugs directly against unpatched main with a standalone script using InMemoryBackend/ObjectStore/RichExplorer (no live deepset Cloud credentials needed):

  store = ObjectStore(backend=InMemoryBackend(), ttl=0)
  obj_id = store.put(["first", "second", "third"])
  RichExplorer(store).explore(obj_id=obj_id, path="[0]")
  # ValueError: Object 'obj_001' does not have a value at path '[0]'.

• Added 3 new tests to test/unit/tokonomics/test_explorer.py: test_parse_path_lone_bracket_index, test_explore_lone_bracket_index_on_list, test_parse_path_empty_brackets. Confirmed all 3 fail against unpatched explorer.py (git stash) with the exact IndexError/ValueError described above, and pass after the fix.
• Updated test_parse_path_simple, which previously asserted _parse_path("attr") == "attr" (a bare string) — that assertion encoded the old, inconsistent contract. Updated it to assert the now-uniform Path return type and verify it still navigates correctly via glom.
• uv run --dev pytest -m "not integration" — 803 passed, 4 skipped (pre-existing, unrelated), no regressions.
• make lint / make format-check — clean.
• make types (mypy) — clean, 146 source files.

Notes for the reviewer

Small, behavior-preserving fix — no change for existing multi-part paths (already used Path(*parts)); only the previously-broken single-part case changes, from a crash/false-negative to correct navigation.