fix(cli): preserve run_id, validate JSON shapes, clamp quality_score

[lightzt99]

Summary

Five correctness fixes in the CLI paths for memory import/export, the interactive shell, the quality command, and JSON-shaped CLI options.

Fixes

1. run_id dropped on import round trip

Memory.import_memories() forwarded messages/user_id/agent_id/metadata but not run_id. Session-scoped memories lost their run scope after a JSON/CSV export+import.

• src/powermem/core/memory.py: pass run_id=memory.get('run_id') to self.add()
• Regression tests: JSON + CSV round-trip asserts kwargs["run_id"] is preserved

2. JSON import accepts invalid top-level shapes

{"content": "one"} (a dict, not the supported array) decoded successfully and imported 0 succeeded, 0 failed with exit code 0.

• src/powermem/utils/io.py: import_from_json() now raises ValueError on non-list top-level shapes
• CLI import already catches Exception and exits non-zero
• Tests: unit test_import_json_dict_top_level_rejected, CLI test_import_dict_top_level_fails

3. Interactive shell dropped --run-id on add/search/list/export

_cmd_export/_cmd_add/_cmd_search/_cmd_list all silently lost run_id.

• src/powermem/cli/commands/interactive.py: all four now forward run_id (long --run-id and short -r) to the underlying SDK call. run_id=True (flag with no value) collapses to None.
• Help text updated to advertise --run-id
• Tests: TestInteractiveExportRunId (3) + TestInteractiveAddSearchListRunId (6) covering with/without run_id for all four commands

4. quality_score could go negative

Empty memories were counted as both empty and short (len 0 < 10), yielding quality_score: -1.0 for a single empty memory.

• src/powermem/cli/commands/memory.py: short_count now requires 0 < len(...) < 10 so empty rows aren't double-penalized. Score also clamped to [0, 1] as a backstop.
• Test: test_quality_score_non_negative_for_empty_only asserts 0.0 ≤ quality_score ≤ 1.0 for a single empty memory

5. CLI --metadata / --filters accepted any JSON shape

add/search/list/update parsed the value with json.loads and passed it straight through. '[]' or '"str"' would flow downstream as the wrong type with confusing behavior.

• src/powermem/cli/commands/memory.py: new _parse_json_dict() requires a JSON object and errors out otherwise (Invalid {field}: expected JSON object, got array)
• Tests: TestMetadataFiltersShapeValidation — dict accepted, array/string rejected for add --metadata, search --filters, list --filters

Trailing whitespace

git diff --check upstream/main...HEAD is clean across the full PR diff.

Test plan

• pytest tests/unit/test_cli_export_import.py tests/unit/test_cli_profile.py tests/unit/test_cli_config_show.py → 75 passed (12 new)
• pytest tests/unit -q -k "io or import or export or quality or interactive" → 277 passed
• git diff --check upstream/main...HEAD exit 0

Files

• src/powermem/core/memory.py — import_memories forwards run_id
• src/powermem/utils/io.py — import_from_json validates list top-level
• src/powermem/cli/commands/interactive.py — run_id forwarded in add/search/list/export; help text
• src/powermem/cli/commands/memory.py — _parse_json_dict for metadata/filters; quality_score clamped + non-double-penalized short count
• tests/unit/test_cli_export_import.py — 12 new regression tests

[wayyoungboy]

Requesting changes because interactive mode still drops the session scope on the profile paths.

The PR now forwards --run-id for plain interactive add, search, list, and export, but add --with-profile --run-id ... still calls ctx.user_memory.add(...) without run_id, and search --add-profile --run-id ... still calls ctx.user_memory.search(...) without run_id. Both UserMemory.add() and UserMemory.search() support run_id, and the non-interactive CLI already forwards it in these profile modes.

I verified the current targeted CLI unit tests pass, but a minimal mocked interactive session still observes run_id=None for both profile calls. Please forward run_id in those two branches and add regression coverage for --with-profile --run-id and --add-profile --run-id.

[wayyoungboy]

Thanks for the PR. I do not think this is ready to merge yet because the previous run-id blocker still needs to be addressed.

Please make sure the interactive profile paths preserve the session scope too:

• add --with-profile --run-id ... should pass run_id into UserMemory.add().
• search --add-profile --run-id ... should pass run_id into UserMemory.search().

Please also add regression coverage for both profile-mode paths. Once those are fixed, this should be ready for re-review.