Fix positional put smart type detection

[arcaputo3]

Summary

• route positional put values through the same currency, percent, ISO-date, number, and boolean detection used by batch put
• add --no-detect for literal-text parity with batch detect:false, including the streaming path
• preserve existing cell styling when streaming detected formats and reuse generated style IDs across ranges
• reject unparseable strings for explicit date, datetime, and time batch formats instead of writing text with a date-like number format
• update CLI, bundled-skill, and 1904-streaming limitation documentation

Root cause

Positional put used ValueParser.parseValue, which only recognized numbers and booleans. ISO date strings therefore became text even though batch put already used FormattedParsers.detect. The explicit date-format path also fell back to a text value while retaining the requested number format, producing cells that looked like dates but broke date formulas.

Impact

Positional ISO dates now land as numeric Excel dates with NumFmt.Date, so downstream formulas such as XIRR and EOMONTH receive typed values. Currency and percent formatting also matches batch behavior. Detection remains intentionally ISO-date-only; use --no-detect when date-like or numeric input must remain text.

The streaming implementation preserves existing font/fill/border/alignment and existing non-General number formats. Formatted range fills share generated cellXf entries instead of creating one duplicate style per cell.

Validation

• ./mill --no-server xl-cli.checkFormat
• ./mill --no-server xl-cli.test
• git diff --check
• regression coverage for CLI option parsing, in-memory and streaming date writes, literal opt-out, style preservation, bounded streaming style growth, and invalid explicit date formats

Closes #431

[claude]

Review: Fix positional put smart type detection (#431)

Reviewed the full diff plus the surrounding code paths (FormattedParsers.detect, Sheet.putSingle/mergeStyles, StylePatcher.addStyle, StreamingTransform, and readExistingWorksheetMetadata). This is a well-scoped, high-quality change: the fix reuses the existing FormattedParsers.detect rather than reimplementing detection, the streaming style-reuse logic faithfully mirrors the in-memory Sheet.put(Formatted) semantics, the docs/skill are updated in lockstep, and the tests cover the important round-trips. Nice work.

A few things worth considering before merge:

Correctness / behavior

1. Explicit time format can no longer accept a time-of-day string (minor regression). In BatchParser, the explicit date | datetime | time branch now hard-errors when FormattedParsers.parseDate fails. But parseDate only accepts ISO YYYY-MM-DD (it does LocalDate.parse and returns midnight), so an input like {"op":"put","format":"time","value":"14:30"} previously fell back to Text and now fails the whole batch. The stricter rejection is clearly right for date/datetime, but for time it means the only accepted input is a bare date (which renders as 00:00), so legitimate-looking time strings now error. Worth either documenting that time requires a date input or leaving time with the softer fallback.

2. Breaking behavior change for positional put (intended, but flag for release notes). Values like 50%, $1,234.56, and 2024-03-15 that previously landed as text now become typed numbers/dates. This is the point of the fix and --no-detect is a clean opt-out, but it's a semantics change for existing scripts (e.g. codes/IDs that look like ISO dates) and should be called out in the changelog.

Performance

3. Formatted streaming puts now make two full worksheet passes. For a formatted --stream put, scanExistingStyles walks the entire worksheet to recover the target cell's existing style id, and then transformWithMetadata walks it again to apply the patch. Memory stays O(1), but read time roughly doubles on very large files. Since --stream exists specifically for huge files, it may be worth folding the existing-style lookup into the transform pass later. Not a regression (formatted streaming puts weren't supported before), just a note. Plain-value puts correctly short-circuit (readExistingWorksheetMetadata returns empty when no col/row ops, and scanExistingStyles is skipped when there are no formatted refs), so the common path is unaffected.

Test coverage

4. The "preserve existing explicit format" branch is untested. The streaming tests exercise the merge branch (existing General + bold, keep bold, apply Date), the range-dedup path, and --no-detect, but nothing covers the else (existingStyle, existingStyleId) branch where the target cell already has a non-General numFmt (e.g. a Currency-formatted cell receiving a date value should keep Currency and reuse its style id). A test there would lock in the intended Sheet.put(Formatted) parity.

5. Positional-put detection is only tested for ISO dates. Currency, percent, and boolean detection for positional put rely on FormattedParsers.detect (which has its own tests), but a one-line assertion each for put A1 "\$100" and put A1 "50%" would guard against future regressions in the wiring.

Nit

6. In ValueParser.parsePutValue, the case Formatted(CellValue.Text(_), numFmt) reuses numFmt from detect, but detect only ever returns NumFmt.General for Text. Using NumFmt.General directly would make the intent clearer and avoid implying a text value could carry a non-General format.

Overall this looks solid and ready to merge once the time-format edge case (#1) is resolved or documented; the rest are follow-up-friendly.