fix(api-validation): dot-gated extension extraction so .bash_history sorts as empty

[githubrobbi]

Symptom

T62 --sort extension asc fails on Windows live MFT with:

Error: Not asc: bash_history > 2008
RPC: search({"limit":10,"pattern":"*.*","sort":"extension"})

CLI lane (uffs.exe "*.*" --sort extension --limit 10) returns the correct order — .bash_history first with empty Extension "", then digit-prefixed "2008", "BIN", "bin", "csv", "dat". MCP lane is also green (fixed in #69 / v0.5.74). Only the pure-RPC API validation lane fails.

Root cause

scripts/windows/api-validation.rs derives the Extension column locally from the daemon's name JSON field via rpc_field_computed("_ext"):

"_ext" => {
    let name = field_str(row, "name");
    let ext = name.rsplit('.').next().unwrap_or("").to_lowercase();
    Some(ext)
}

".bash_history".rsplit('.').next() returns "bash_history" (post-dot suffix when the dot is at index 0), instead of the empty string the sort engine uses. Same naive extractor at line 2002 of the type_* allowlist validator: dotless names get their full lowercased name as a phantom extension.

This was the third copy of the bug. CLI/format and MCP lanes were fixed in v0.5.74 (#69) by routing through extension_from_name / extract_extension_after_dot. The API validation lane was missed because it has its own validator-side extraction (no shared helper with the workspace).

Fix

Single helper, two call sites:

fn extract_ext_dot_gated(name: &str) -> String {
    let Some(dot) = name.rfind('.') else { return String::new(); };
    if dot == 0 || dot + 1 >= name.len() {
        return String::new();
    }
    name.get(dot + 1..).unwrap_or("").to_ascii_lowercase()
}

Mirrors extract_extension_after_dot in crates/uffs-core/src/search/filters/ext_match.rs. Returns "" for: dotless names (README), leading-dot hidden files (.bash_history), trailing-dot names (foo.).

Used in:

• _ext computed column → fixes T62 sort + Extension column display in API validator.
• type_* allowlist (type_code, type_document, type_executable, type_picture, type_system) → no longer mis-classifies dotless rows by their full name.

Verification

• rust-script --package + cargo check on the generated package: clean compile, no new warnings.
• lint-fast (file-size, fmt-check, lint-prod, lint-tests, lint-ci, typos, reuse): all green.
• lint-pre-push (incl. smoke + check-windows): all green.
• Equivalent dot-gated semantics already pinned by regression tests in crates/uffs-format/src/writer.rs, crates/uffs-core/src/output/tests.rs, and crates/uffs-mcp/src/tools/search.rs (fix(mcp): dot-gated extension extraction so .bash_history sorts as empty #69).

Out of scope

scripts/windows/api-validation.rs is a rust-script (not a workspace crate), so it is not gated by just lint-ci. Pre-existing clippy nits in unrelated parts of the file (unnecessary_filter_map, unnecessary_map_or, etc., lines 2689 / 2862 / …) are left untouched.

Affects

• ✅ T62 --sort extension asc (the originally reported failure)
• ✅ type_code / type_document / type_executable / type_picture / type_system validators

Closes the third copy of the dot-gated extension bug; CLI / MCP / API lanes now share semantics.