fix(get): accurate error when a glob excludes all tracked files under a path

[CGMossa]

TODO: replace this line with your own framing, in your own voice.

AI-written details

Summary

• resolve_paths_for_get reported "The following paths are not tracked by DVS" for every explicit path that matched no tracked file, even when the path has tracked files and the provided glob merely excluded all of them (e.g. dvs get data/ --glob '*.parquet' where data/ holds only tracked CSVs).
• When a glob is provided and a path matches nothing, the path is now re-checked without the glob. If tracked files exist under it, the error reads "The following paths have no tracked files matching glob ''". Otherwise the existing "not tracked by DVS" message stands.
• Mixed failures list each path under the correct heading. Both cases still refuse the whole batch, exactly as before.
• specs.md: the Globbing section now states that explicit file paths bypass the glob in add, get, and status. The get section documents the glob-excluded error variant.
• Scope: only resolve_paths_for_get in dvs/src/globbing.rs. PathFilter::matches, resolve_paths_for_add, get.rs, and add.rs are untouched.

Test plan

• cargo test -p dvs (81 passed, includes three new tests: glob excludes everything, untracked path with glob, mixed failure with both headings)
• cargo clippy -p dvs --all-targets (no warnings)
• cargo fmt --check
• Manual CLI check: tracked dir + excluding glob names the glob, untracked path + glob still says "not tracked", mixed run shows both headings, explicit file still bypasses the glob (all exit 1 except the bypass case)

Notes

• The exact-path-bypasses-glob behavior of PathFilter::matches is deliberately unchanged (covered by get_explicit_file_ignores_glob) and is now stated in the spec for all three commands.
• specs.md edits are two small hunks (Globbing section, get CLI error paragraph). Trivial conflicts possible with fix(core): surface bare directory in add instead of silent drop (closes #225) #239, docs(spec): get verifies a temp file and never deletes the local target #256, and fix(core): never delete storage blobs on rollback #257, which touch nearby lines.
• The "check rpkg" CI job can fail on a stale vendor cache for path-dep'd core changes. That is a known cache-key issue, not caused by this PR.

---

Drafted by Claude (claude-fable-5). Reviewed by the author.

[CGMossa]

This is related to #277. So we have to think about where this makes the most sense to add ignored paths.

[CGMossa]

TODO: replace this line with your own framing, in your own voice.

AI-written details: orthogonal to #277, composes cleanly with it, where ignored paths should live

Summary

• Verified on the branch: 84 unit tests pass, clippy and fmt clean. The new tests pin all three outcomes (glob excluded, untracked, mixed batch with both headings in order).
• Direct answer to the question in this thread (where should ignored paths live): this PR and fix(core): ignore reserved paths in tracked set and reject them in add validation #277 are orthogonal, and neither should absorb the other. fix(core): ignore reserved paths in tracked set and reject them in add validation #277 defines what the tracked set contains, at the DvsPaths::tracked_paths() choke point in paths.rs (its globbing.rs diff is test-only, so the file overlap is proximity, not coupling). This PR never defines what is ignored. It only explains why an explicit path resolved to nothing, given whatever tracked set it was handed. A per-invocation --glob is transient user input, not a property of the repo, so it does not belong in a shared reserved-path registry. The real unification target is fix(add): glob walk skips .git, dvs.toml, .gitignore, and broken symlinks #265 plus fix(core): ignore reserved paths in tracked set and reject them in add validation #277: fix(add): glob walk skips .git, dvs.toml, .gitignore, and broken symlinks #265's .git walk pruning should reuse fix(core): ignore reserved paths in tracked set and reject them in add validation #277's is_reserved_path instead of a second hardcoded encoding of the same rule (details on the fix(core): ignore reserved paths in tracked set and reject them in add validation #277 thread).
• The two compose correctly with no rework. I merged fix(core): ignore reserved paths in tracked set and reject them in add validation #277 into this branch locally (clean auto-merge) and the combined tree builds, passes all 93 tests, and is clippy clean. Composition also yields the right message: in a poisoned repo, dvs get .git/refs --glob '*' reports "not tracked by DVS" rather than blaming the glob, because fix(core): ignore reserved paths in tracked set and reject them in add validation #277 filters reserved sidecars out of the tracked set upstream of this PR's re-check. That is the desired outcome and it falls out for free.
• The stale base is harmless. This branch sits on ac72ed8, a few commits behind main, but the files it touches are byte-identical between its merge-base and current origin/main, so GitHub's MERGEABLE is trustworthy and no rebase is required.
• The re-check has a coherent definition worth stating explicitly: a path is blamed on the glob only when the same invocation without --glob would have found tracked files. One boundary case follows from PathFilter::matches: dvs get data --glob '**/*.parquet' without -r, where data has tracked files only in subdirectories, reports "not tracked" because the no-glob re-check honors the non-recursive direct-child rule while ** globs reach any depth. That matches what plain dvs get data would say, so I read it as consistent behavior, not a bug.

Notes

• just check-std-fs fails on this branch, but the hits are identical to the failure already present on origin/main (test-module imports plus two call sites in add.rs and get.rs). Not introduced here, worth a separate cleanup.
• Merge order: independent of fix(core): ignore reserved paths in tracked set and reject them in add validation #277 and fix(add): glob walk skips .git, dvs.toml, .gitignore, and broken symlinks #265, can land at any point relative to them.
• The specs.md additions comply with the house style.

---

Drafted by Claude (claude-fable-5). Reviewed by the author.