fill: review-driven cleanup, fix MapSelected unsorted-selection bug, ~22% faster hot path

[jqnatividad]

Summary

• Correctness fix: MapSelected previously assumed selection indices were sorted and unique, so user-supplied selections like qsv fill -- "3,1" file.csv or qsv fill -- "1,1" file.csv silently dropped columns from the fill. Sort + dedup at construction; added regression tests.
• Doc/UX: USAGE now states that --default takes precedence over --first / forward-fill (previously silent).
• Hot-path perf (~22% faster): filled rows are now written directly into a reusable csv::ByteRecord via a new fill_into() method, eliminating two Vec<Vec<u8>> allocations per row on the common no---backfill path. Grouper lookup uses get_mut() so key.clone() and the default-value clone only happen on insertion of a new group.
• Smaller cleanups: // safety: comment on the final-buffer-drain unwrap(), as_ref().or_else().cloned() in GroupValues::fill, hoisted default_value clone into or_insert_with, is_empty() instead of == b"", hoisted the (default, first) match into a MemorizeKind enum computed once outside the loop.

Benchmark

NYC_311_SR_2010-2020-sample-1M.csv (1M rows, 514 MB, 3 selected columns with high empty rates), 5 runs each via hyperfine:

Mode	Baseline	This PR	Speedup
forward fill	1.829 s	1.429 s	1.28×
--groupby fill	1.870 s	1.456 s	1.28×
--default fill	1.758 s	1.372 s	1.28×
--backfill	1.815 s	1.837 s	within noise (slow path unchanged)

Output is bit-identical to baseline (md5 verified for forward + groupby modes on the full 1M-row file).

Test plan

• All existing fill tests pass (cargo t fill -F all_features)
• Three new regression tests added:
  • fill_forward_unsorted_selection — covers MapSelected ordering bug
  • fill_forward_duplicate_selection — covers dedup behavior
  • fill_default_overrides_first — locks in the documented precedence
• cargo +nightly clippy -F all_features -- -W clippy::perf — clean
• cargo +nightly fmt — applied
• docs/help/fill.md regenerated via qsv --generate-help-md
• Manual md5 comparison of full output vs baseline (forward + groupby)

🤖 Generated with Claude Code

[codacy-production]

Up to standards ✅

🟢 Issues 0 issues

Results:
0 new issues

View in Codacy

🟢 Metrics 7 complexity

Metric	Results
Complexity	7

View in Codacy

_{NEW Get contextual insights on your PRs based on Codacy's metrics, along with PR and Jira context, without leaving GitHub. Enable AI reviewer
TIP This summary will be updated as you push new changes.}

[Copilot]

Pull request overview

This PR improves the fill command’s correctness, clarifies option precedence in the help text, and optimizes the non---backfill hot path by avoiding per-row heap allocations.

Changes:

• Fixes a correctness bug where MapSelected could mis-handle unsorted or duplicate selection indices by sorting + deduping at construction time (with regression tests).
• Documents that --default takes precedence over forward-fill / --first (making them no-ops when default is set).
• Optimizes the no---backfill path by filling directly into a reusable csv::ByteRecord and reducing cloning on group insertion.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File	Description
src/cmd/fill.rs	Fixes selection-order/dup handling, adds fill_into() fast path, and updates USAGE to document --default precedence.
tests/test_fill.rs	Adds regression tests for unsorted/duplicate selection and --default precedence over --first.
docs/help/fill.md	Regenerates help text to reflect --default precedence.