Rework issue #451: draggable table column resizing, made ephemeral

[schuyler]

Draggable table column resizing in the live preview, reworked from @aseelye's PR #451 contribution to hold width state in memory only, with no document mutation or persistence of any kind.

Behavior

• Drag a column-boundary handle in a rendered table to resize it live in the preview.
• Double-click a handle to reset that column to natural sizing.
• Widths survive incremental preview updates (typing) because the preview's JS context persists across DOM replacement.
• Widths reset on a full preview reload (style change, window reopen) and on document close. Nothing is written to the document or to disk.
• Export (HTML/PDF) is unaffected by resize state.

What changed from @aseelye's original PR #451

His interactive front end (drag handles, table identity via header-hash + index, minimum column width) is unchanged. Removed entirely: the native persistence layer — bridge token, x-macdown-table-layout:// custom URL scheme, table-layouts.json sidecar file, and the associated load/save/reset logic in MPDocument.m/MPRenderer.m/.h. table-resize.js now holds width state in a closure-scoped map instead of round-tripping through the native bridge.

His authorship is preserved via a straight cherry-pick (a5f9b40) as the base commit, with the persistence removal as a separate commit on top.

Judgment calls

• No Markdown parsing, no export of resized widths. The feature is purely DOM-level and is unaffected by the hoedown → cmark-gfm parser swap (Address issue #77: migrate Markdown engine from hoedown to cmark-gfm #515 / v3000.1).
• Table identity is header-hash + index, unchanged from the original PR. Renaming a header or reordering tables (which shifts the index) loses that table's widths for the session — accepted as an ephemeral-view-state tradeoff.
• export.css's bundled th { white-space: nowrap } change rides along unchanged from the original PR. It affects all table rendering, including export, not just the live-resize feature — a UX change serving the same table-readability goal, kept as part of this PR rather than split out.
• In-document persistence remains a future feature, tracked separately against a dash-count design built for cmark-gfm rather than the sidecar-file approach removed here.

Testing

CI=true xcodebuild test -workspace "MacDown 3000.xcworkspace" -scheme "MacDown (MacDown 3000 project)" -destination 'platform=macOS': 1225 tests, 76 expected failures, 0 unexpected.

Headless tests cover: the renderer emits no layout JSON or table-layout bridge token; table-resize.js is registered in the preview scripts; export HTML/CSS behavior is unaffected; the removed URL-scheme handler leaves no trace. Drag interaction and width re-application after DOM replacement run only in a live WebView and are manual-QA items.

Related to #451

[schuyler]

Local test run against this branch (three separate invocations — warm DerivedData, fully cold DerivedData, and reproducing CI's exact scheme name and xcodebuild flags) passes 1303 tests: 76 expected XCTExpectFailure failures, 0 unexpected.

The Tests workflow failures above (macos-14 cancelled, macos-26 failed) are the intermittent CI-runner hang under separate investigation — both matrix jobs stall for ~2 minutes on unrelated tests with no local reproduction across three attempts, consistent with a runner-level issue rather than a regression in this branch.