Port the shared table engine so the core matches JetBrains and Notepad++

[krotname]

What this is

An independent audit of the VS Code port found that src/core.ts was an independent
reimplementation of the shared Markdown table engine rather than a port of it, and that it
diverged from the reference on roughly 10% of a differential corpus — including several
divergences that corrupt documents.

This replaces the core with a faithful port of MarkdownTableEngine.java (mirrored by the
Notepad++ MarkdownTableCore.cpp), including the shared Unicode width tables.

How the divergences were found

A differential harness runs the same operations through the reference core
(markdown-table-core-0.3.1.jar) and through this core, comparing every field of EditResult.

corpus	cases	before	after
hand-written edge cases + fuzz	11 719	8 048 differing	0
width probes, adversarial separators, CSV, deep Unicode	19 089	not run	0
chained operation sequences (up to 6 steps each)	6 000	not run	0

Defects fixed

• Prose was pulled into the table and rewritten. tableRangeEnd now requires continuation
  rows to open with a pipe when the header and separator do, so a line such as tail | prose
  after a table is left alone. Previously any following line containing a pipe was absorbed into
  the table range and reformatted.
• Cells were silently dropped. parseTable now takes the column count from the widest row and
  pads short rows. It previously truncated every row to the separator's width, so a row with an
  unescaped pipe (| x|y | 0 |) lost its trailing cells on any edit.
• Delete Row could destroy the table. canDeleteRow refuses to delete the row above the
  separator, so deleting the header no longer leaves a separator-first block that is not a table.
• The caret landed past the content in right and centre aligned columns. targetColumnOffset
  is now the character offset of the cell content inside the formatted line, matching the
  reference. It previously carried the padding width, which was then added to a position that had
  already skipped that padding — an overshoot equal to the padding. cellCharacterOffset is no
  longer needed and is removed.
• Valid separator rows were rejected. isPotentialSeparatorLine accepts spaced dashes
  (| -- - |), = rules (|===|) and short |--- lines. Conversely, resolveTable now rejects
  a pipe block that has no separator row at all.
• Fitting was not idempotent. applyWrappedToWidth and the manual resize actions unwrap
  continuation rows first and use the reference budget algorithm. Narrow and widen follow the
  content width instead of a fixed width-3 floor.
• Conversion and creation differed. fromDelimited keeps cell text verbatim instead of
  collapsing internal whitespace, newTable emits Column N headers and puts the caret on the
  first data row, and both report the shared status messages.
• Display widths depended on the host runtime. Widths now come from the same static tables the
  other cores use, instead of Intl.Segmenter plus \p{Extended_Pictographic}, so measurements no
  longer drift with the Node/ICU version.

Verification

• npm run typecheck — clean
• npm run test:unit — 61 passing (43 existing golden/contract tests unchanged, 18 new)
• npm run test:coverage — core.ts 97.62% lines, 85.66% branches, 99.31% functions
• npm run build, npm run package, npm run test:vsix — VSIX installs and enumerates
• npm run test:e2e — 4 passing in the real Extension Host, including a caret-position check on a
  right aligned cell and a guard that prose after the table is untouched
• npm audit — 0 vulnerabilities
• 10 000-row table: ALIGN 39 ms, SORT 24 ms, FIT 56 ms (was ~500 ms)

The golden fixture file is unchanged and still matches the JetBrains copy byte for byte.

🤖 Generated with Claude Code

https://claude.ai/code/session_01KsNFrStLUQaEZNtCybK4XD

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 2d717b4474

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".