M3X-1480: Fix tables redlining and refactor the package

[ragoolkr]

Ticket

Tables: tables are now compared as whole structures. Each old/new table pair becomes one merged table, whole rows and columns are marked as added or deleted, merged cells are handled, and each cell content is diffed through diffCore word level diff.

Refactoring: the single js/htmldiff.js is split into modules. src/core/ holds the existing word level diff (tokens, matching, operations, rendering) and src/tables/ holds the new table code, while the public API (export = diff) stays the same.

TypeScript: the library and tests are now written in TS, built from src/ into js/ with generated .d.ts files (so js/ is no longer committed). Tests run through mocha + ts-node, and publishing does a clean build plus the CLI build.

Rich text tables

rich-text-tables.mov

Step list table

step-list.mov

Document section table

document-section.mov

Caution

Still need to clean up client and small css update and for document sections need to add a html-diff-id to cells so algo understands which is actually the real primary item ref as multiple item refs are involved in many section.

[sirenevenkii]

didn't looked at the actual logic yet.

[sirenevenkii]

I didn't check all the code, but I made some sanity checks using AI. I think we're good. great job overall 👍

one thing that worth looking at is performance. I did some testing with AI and it seems like there are cases where it behaves suboptimal. this PR is big enough and I think the goal is achieved, but we need to have a look at it separately.

---

Performance results (AI response as is)

The PR's speed on tables depends on how heavily they were edited: it's 3–4× slower on large, lightly edited or unchanged tables and 3–7× faster on heavily edited ones. Content without tables runs at the same speed as on develop. The slow path scales quadratically with row count.

Workload	Size	develop	PR	PR vs develop
Text, 5% of words edited	5 / 51 / 505 KB	0.66 / 21.3 / 1999 ms	0.75 / 21.0 / 1878 ms	same (0.94–1.13×)
Text, identical	505 KB	~0 ms	4.9 ms	scan overhead only
Table 10×5, light edits	4 KB	0.46 ms	1.7 ms	3.7× slower
Table 50×6, 10% cells + a row added/deleted	21 KB	7.3 ms	24.7 ms	3.4× slower
Table 200×6, 10% cells + a row added/deleted	82 KB	96 ms	301 ms	3.1× slower
Table 200×6, cell ids on rows, 10% cells	92 KB	78 ms	294 ms	3.8× slower
Table 200×6, identical	82 KB	~0 ms	295 ms	develop returns at once
Table 200×6, 60% cells edited	82 KB	182 ms	30 ms	6× faster
Table 300×8, long cells, 60% edited	403 KB	1483 ms	218 ms	6.8× faster
20 sections, each with a 10×4 table, all edited	66 KB	56 ms	52 ms	same

Scaling with row count, for a table with 6 columns where 5% of cells changed:

Rows	50	100	200	400	800
develop	7 ms	22 ms	110 ms	368 ms	1.8 s
PR	23 ms	83 ms	295 ms	1.16 s	4.5 s
PR, identical table	25 ms	78 ms	289 ms	1.12 s	4.3 s