fix(uffs-core): drop per-row format!() allocations in path-resolver + result-conversion hot paths (Phase 6d, refs #193)

[githubrobbi]

Phase 6 sub-phase 6d deliverable per phase_6_ownership_borrowing_allocation_implementation_plan.md §5d + §3.4 (local plan, gitignored under /docs/dev/).

What this changes

Two surgical hot-path category-δ fixes inside uffs-core:

Site	File	Refactor
Per-row path assembly	crates/uffs-core/src/path_resolver/fast.rs:add_path_column_with_dir_suffix	Replace two per-row format!("{parent_path}{file_name}") / format!("{parent_path}\\{file_name}") with in-place push_str on the already-owned parent_path: String
Per-row drive label	crates/uffs-core/src/search/dataframe_convert.rs:display_rows_to_dataframe	Replace per-row format!("{}:", row.drive) (Vec<String>) with a static [&str; 26] table indexed by DriveLetter::alphabet_index() (Vec<&'static str>)

Both run inside per-row map closures over search-result rows; each row previously incurred an unconditional small String allocation that the new code avoids.

Why this is the right shape

Phase 6d — hot-path format! / to_string audit — sweeps uffs-mft::io::parser, uffs-mft::parse, uffs-core::search, uffs-core::path_resolver, and uffs-core::aggregate for the per-record / per-row formatting hot spots flagged in plan §1.2 (HIGH priority).

The audit produces a small surgical-fix surface:

• uffs-mft::io::parser — zero format! / .to_string() calls in the parse hot path (columns.rs, merger.rs, zero_alloc.rs, dataframe_build.rs). The 205 format!() count from the 6a baseline lives in cold paths (parse-error variants, log messages, USN status, IOCP error context). Plan §0.2 risk-register row predicted this.
• uffs-core::path_resolver::fast.rs — two per-row format! calls in add_path_column_with_dir_suffix, fixed by this PR.
• uffs-core::search::dataframe_convert.rs — one per-row format! for drive label, fixed by this PR.
• uffs-core::path_resolver::legacy.rs:104 — format!("{}:\\{}", drive, components.join("\\")) is irreducible (must allocate one owned String result per resolve); cat-β KEEP.
• uffs-core::search::query::mod.rs, backend.rs — pattern.to_owned() / needle.to_owned() sites are once-per-query (query prep), cat-β KEEP. Candidate for §6e Cow<'_, str> expansion.
• uffs-core::aggregate::finalize.rs:675-705 — format_field called per (bucket × sample size), not per record. Aggregation framework only samples top-K records per bucket. Marginal cat-γ, KEEP.
• <UNKNOWN_0x...> / <DELETED:...> / <EXT:...> / <CORRUPT:...> placeholders in parse/{direct_index,direct_index_extension,forensic/base,forensic/extension,forensic,full,placeholders}.rs — conditional formatting that fires only on unknown / deleted / extension / corrupt / placeholder records. None fire on every record in the parse hot path. Cat-γ KEEP.

Verification

• just lint-pre-push ✅ (96s): every gate green — fmt, file-size, gates-drift, hooks-drift, workflow-drift, fast-drift, manifest-drift, commit-subjects, vet, vet-audit-discipline, machete, typos, reuse, cargo-check, lint-ci, lint-prod, lint-tests, rustdoc, doc-tests, tests, smoke, deny, lint-ci-windows.
• cargo nextest run -p uffs-core ✅ 817 passed, 3 skipped, 0 failed — same numbers as pre-fix.
• Affected test paths exercised: add_path_column_with_dir_suffix_*, add_path_column_*, add_path_column_parallel_* (path-resolver suite); display_rows_to_dataframe_*, results_to_dataframe_* (result-conversion suite). All pass — observable behavior is byte-identical.

Behavior & contracts

• Public API unchanged: add_path_column_with_dir_suffix and display_rows_to_dataframe keep their existing signatures and return types.
• Observable behavior preserved:
  • Paths render with the same separator handling (test corpus covers parent_path.ends_with('\\') true and false branches, plus the file_name == "." root-of-volume case).
  • Drive labels render exactly as <letter>: for A..Z — the previous format! produced identical output for the construction-valid range.
• For the unreachable fallback path (a hypothetical out-of-range DriveLetter, which cannot occur per parse() invariants), the new code yields "?:" versus the previous code's <letter>:. Since this branch is unreachable by construction, the user-visible difference is zero.

Strict-lint posture

• No suppression hacks introduced. Both edits replace formatting allocations with idiomatic Rust patterns (push_str mutation; static-table lookup via get().copied()).
• The .unwrap_or("?:") fallback in display_rows_to_dataframe satisfies clippy::indexing_slicing = "deny" without needing a per-site #[expect].
• Cross-references to Phase 6d sub-phase added in inline doc comments.

Allocation savings

Per N-row search result (typical N = 1k–100k):

• Site 1 (add_path_column_with_dir_suffix): N small String allocations + N drops (replaced by N in-place push_str operations that reuse the parent_path buffer).
• Site 2 (display_rows_to_dataframe): N tiny (2-byte) String allocations + N drops + 1 Vec<String> allocation (replaced by 1 Vec<&str> allocation; labels live in rodata).

Wall-clock impact will be quantified by sub-phase 6g (bench refresh on mft_read, query, search_benchmarks).

refs #193