Fix table cell font src inheriting an unrelated font family

[MahathirMohammadShuvo]

Fixes #1746.

The bug

A table resolves each cell's font by merging the column, row and cell styles key
by key:

const font = deepMerge({}, colStyle.font, rowStyle.font, cell.font);

src and family do not survive that treatment, because they are not
independent settings: family names a subfamily/variation inside the given
src. When a row style supplies both and a cell overrides only src, the
row's family is carried over and doc.font(src, family) is handed a pair that
was never configured.

Two ways that goes wrong, both from the issue:

doc.table({ rowStyles: [{ font: { src: SOME_TTC_FONT, family: FONT_FAMILY } }] })
   .row([{ text: 'Hello World', font: { src: SOME_TTF_FONT } }]);
// Error: Variations require a font with the fvar, gvar and glyf, or CFF2 tables

doc.table({ rowStyles: [{ font: { src: SOME_TTC_FONT, family: FONT_FAMILY } }] })
   .row([{ text: 'foo' }, { text: 'bar', font: { src: SOME_TTF_FONT } }]);
// no error, but SOME_TTF_FONT is silently ignored - FONT_FAMILY is already
// cached against the first font, so the cell renders in the wrong typeface

Setting a font for a row and overriding it on one cell is ordinary use, and the
silent variant is the worse of the two: the document renders, just with the
wrong font.

The fix

Resolve the font by walking the levels from least to most specific and treating
src and family as one unit:

• a level that sets src also resets family, so a stale family cannot
  outlive the font it belonged to;
• a level that sets only family still refines the inherited src, so
  selecting a variation from a row or column font keeps working;
• size is genuinely independent and continues to merge on its own.

Testing

Two tests added to tests/unit/table.spec.js, using the fonts already in the
repo. The first fails without the change with the exact error from the issue:

expected [Function] to not throw an error but
'Error: Variations require a font with the fvar, gvar and glyf, or CFF2 tables.' was thrown

The second covers the case that must keep working - a cell that overrides only
family still refines the row's src.

Full unit suite passes: 387/387. eslint and prettier clean.

[blikblum]

This will fix #1743?

[MahathirMohammadShuvo]

Yes - I tested it both ways rather than assume, and it does.

#1743 is the same line. deepMerge deep-clones every source, so a Buffer or
Uint8Array src came out the other side as a plain object of numeric keys,
which is why it stopped looking like a font. It is also why that merge was slow:
it was walking the whole font file byte by byte.

Same document, tests/fonts/Roboto-Regular.ttf read into a Buffer:

on master        Buffer src      -> Error: Not a supported font format or standard PDF font.
                 Uint8Array src  -> Error: Not a supported font format or standard PDF font.
on this branch   Buffer src      -> ok
                 Uint8Array src  -> ok

Replacing the merge with direct assignment means the src is never cloned, so it
arrives at doc.font() as the same object it went in as.

I have pushed a test covering it, which asserts identity (src === buffer) so a
future reintroduction of cloning fails loudly rather than silently. With
lib/table/normalize.js reverted to master both tests in that block fail:

a cell overriding only src does not inherit the row family
  -> Error: Variations require a font with the fvar, gvar and glyf, or CFF2 tables.
a binary font src is passed through untouched
  -> Error: Not a supported font format or standard PDF font.

Full unit suite passes: 388/388.

Happy to add Fixes #1743 to the description if you would like both closed by
this PR.

[blikblum]

Many thanks.