[LabDennis]

Long ordinary text containing many spaces made date-time detection retry both parsers at every space. CSV imports and cell entry could therefore spend quadratic time deciding that a field was just text.

Closes #1004.

Only the final two token boundaries can begin a supported time (one token, optionally followed by AM/PM). The parser now checks those candidates, preserves the existing ASCII-space delimiter rule, limits date/time component collection to the supported arity, and borrows the lowercased time body instead of copying it again. Unrecognised input retains its original text.

Validation on Windows / Rust 1.95, based on upstream 84acf62:

• All 196 tests across core and XLSX pass, including a CSV regression with a long text field and both workbook date systems.
• A deterministic regression caps candidate count, and 480 combinations cover supported date/time forms and mixed whitespace. The candidate-count regression fails against the original scan.
• Strict core Clippy, formatting, assets, layers and the full WASM gate pass (11 crates; one existing UI dead-code warning).
• Full xtask ci stops at the unchanged collapsible_match finding in crates/xlsx/src/xls/mod.rs:273. Separate parity and keytip checks also report stale generated files, neither changed by this PR. No full-green workspace CI claim.

Bounded microbenchmark, five parses of a synthetic space-rich text field, same machine/profile:

Words	Before	After
1,000	39.46 ms	0.131 ms
2,000	145.75 ms	0.225 ms
4,000	556.19 ms	0.415 ms

This measures input parsing only, not whole-application or complete CSV-import performance. Reproduce with cargo test -p gridcraft-core space_rich_input_timing -- --ignored --nocapture; ordinary tests have no timing thresholds.

Upstream main, issue status and open PRs were rechecked before publication. Related #1132 changes day-first date recognition, #1114 passes workbook date systems to formula arguments, and #1168 adds a conversion preference; this PR keeps the existing grammar and conversion policy.

The branch is now based on upstream d9d09f0baa91. History: 1 → 1 commits. The rewritten file tree was checked against the previous PR plus current upstream, preserving prior integration resolutions.

Scoped Windows regression checks: 18 passed, 1 ignored, zero failures. Formatting, locked Cargo metadata, asset and layer checks passed. This is scoped validation, not a new full-workspace or cross-platform CI claim.

[LabDennis]

Rebased onto current main (f1894c6) as f91ed73; retained the new upstream file-format documentation while keeping the bounded date-time detection fix. Parser and CSV regression tests, WASM, formatting, assets and layers pass. Scoped Clippy still reports two lints in unchanged upstream files (crates/xlsx/src/xls/mod.rs:273 and crates/xlsx/src/tests/passthrough_tests.rs:480); these files are identical to the current target branch and were not changed or suppressed.