Support filter in puma ls

[kerthcet]

What this PR does / why we need it

Which issue(s) this PR fixes

Fixes #

Special notes for your reviewer

Does this PR introduce a user-facing change?

[kerthcet]

/lgtm
/kind feature

[Copilot]

Pull request overview

This PR extends the puma ls command to support optional filtering (pattern + column-based filters), and refactors CLI command implementations into dedicated modules while updating the storage layer to accept filter parameters.

Changes:

• Add puma ls arguments for a regex pattern and -l/--query key-value filters, wiring them through CLI → registry → storage.
• Update SQLite storage + registry APIs to support filtered load_models(...), and rename the model “type” field/column to task.
• Add new CLI modules (ls, inspect, rm) and introduce a new integration test suite.

Reviewed changes

Copilot reviewed 13 out of 14 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
tests/integration_test.rs	Adds end-to-end CLI tests for ls filtering and other commands (currently includes a network-dependent pull).
src/system/system_info.rs	Updates system info to use the new load_models(None) signature.
src/storage/storage_trait.rs	Extends storage trait load_models to accept optional filters.
src/storage/sqlite.rs	Implements SQL filtering, renames type→task in schema/queries, and normalizes stored names/authors to lowercase.
src/registry/model_registry.rs	Propagates new load_models(filters) API and renames r#type→task.
src/lib.rs	Adds minimal crate-level commentary (no functional changes).
src/downloader/huggingface.rs	Updates downloaded model info to populate task.
src/cli/rm.rs	Extracts rm logic into a module with unit tests.
src/cli/mod.rs	Exposes new CLI submodules.
src/cli/ls.rs	Implements ls filtering logic (regex + query parsing) with unit tests.
src/cli/inspect.rs	Extracts inspect logic + display formatting with unit tests.
src/cli/commands.rs	Adds LsArgs, wires new ls/rm/inspect modules, and updates ls table columns.
Cargo.toml	Adds regex dependency.
Cargo.lock	Locks regex dependency.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The initial migration now creates task instead of the previous type column. Existing DBs that already ran the old migration won't be updated by rusqlite_migration, so later queries selecting task will fail. Add a new migration that renames/adds the column (and index) while preserving existing data.

[Copilot]

Lowercasing model.name on write is a backward-incompatible change for existing registries that may have stored mixed-case names previously; combined with lowercasing lookups, those rows become unreachable. If this behavior is desired, add a DB migration to normalize existing rows; otherwise, prefer case-insensitive comparisons (e.g., COLLATE NOCASE) without rewriting stored values.

[Copilot]

Lowercasing the user-provided regex pattern changes the regex semantics (e.g., character classes like [A-Z], escapes, and Unicode properties), not just case-sensitivity. Prefer compiling the original pattern with case-insensitive mode (e.g., RegexBuilder::new(pattern_str).case_insensitive(true)) or lowercasing both sides only when doing plain substring matching.

Suggested change

	let pattern_lower = pattern_str.to_lowercase();
	match regex::Regex::new(&pattern_lower) {
	match regex::RegexBuilder::new(pattern_str)
	.case_insensitive(true)
	.build()
	{

[Copilot]

The ls help text suggests patterns like openai/*, but ls treats this as a regex and openai/* is not a valid regex (it will error because * repeats the preceding token /). Either update the examples to valid regex (e.g., ^openai/.*) or implement glob-style matching if that's the intended UX.

Suggested change

	/// Optional model name pattern to filter (e.g., qwen, openai/*)
	/// Optional model name regex to filter (e.g., qwen, ^openai/.*)

[Copilot]

This test performs a real pull from Hugging Face over the network, which makes CI runs flaky and environment-dependent (network/rate limits/outages). Consider marking it #[ignore] by default, gating behind an env var/feature, or using a mocked/local provider so tests run offline deterministically.