feat(table): add cell wrapping using virt lines

[okuuva]

Continued work on top of #617:

• Rebased on top of the latest main
• Fixed the bottom border rendering when the table was at the end of the file
• Fixed the long header separator rendering
• Fixed the indented wrapped table rendering
• Made wrapping use word boundaries instead of just always wrapping at the character limit
• Fixed the wrapping when continuation markers are in use
• Implemented cell alignment for wrapped cells

[oriori1703]

this also needs to be added to https://github.com/MeanderingProgrammer/render-markdown.nvim/blob/main/lua/render-markdown/types.lua#L286

[andrewroxby]

Been running this branch daily with wrap on — thank you, it makes wide tables genuinely usable. With indent.enabled = true I hit three interactions between the cell wrapping and the indent module, plus a wrap-boundary bug in the slot simulation. All four are fixed on table-wrap-fixes (b9ee242 + c53110f).

1. Overlay-placed visual rows don't include the indent prefix that the virt_lines path adds, so part of an indented table shifts sideways by the indent width.
2. The width budget in compute_wrap_layout doesn't subtract the indent's virtual columns, so under nested headings the right border lands past the window edge.
3. wrapped_slots simulates raw-text wrapping without the indent's inline columns. A row whose raw width straddles the boundary gets a stale slot map — a blank screen line mid-table, with later rows displaced into virt_lines. (The concealed source line still wraps on its raw width; conceal blanks glyphs but doesn't reflow.)
4. wrapped_slots also breaks out of its scan when a screen row fills to exact capacity, but Neovim leaves a following break character hanging at the boundary, so the simulation rewinds the wrap one word early and every later overlay lands one screen row off. Symptoms are width- and content-dependent, which made this one erratic to reproduce.

Happy to open a PR into this branch if useful.