Faster name visitor subset

[behdad]

#3213 (comment)

[anthrotype]

it would be nice if @imran-iq could give this PR branch a try with their go-noto font build to confirm it is fast enough?

[behdad]

Here's a timing I observed:

$ time ./fonttools subset --gids=0-10000 --font-number=0 SourceHanNotoCJK.ttc --timing
Took 0.000s to load font
Took 0.000s to compile glyph list
Took 0.000s to load 'maxp'
Took 0.000s to prune 'maxp'
Took 0.041s to load 'cmap'
Took 0.000s to prune 'cmap'
Took 0.000s to load 'post'
Took 0.000s to prune 'post'
Took 0.000s to load 'CFF '
Took 0.017s to prune 'CFF '
Took 0.000s to load 'GPOS'
Took 0.009s to prune 'GPOS'
Took 0.000s to load 'GSUB'
Took 0.015s to prune 'GSUB'
Took 0.012s to close glyph list over 'cmap'
Took 0.068s to close glyph list over 'GSUB'
Took 1.175s to close glyph list over 'CFF '
Took 0.011s to subset 'cmap'
Took 0.000s to subset 'post'
Took 0.013s to subset 'CFF '
Took 0.000s to subset 'VORG'
Took 0.056s to subset 'hmtx'
Took 0.016s to subset 'vmtx'
Took 0.000s to subset 'GDEF'
Took 0.090s to subset 'GPOS'
Took 0.010s to subset 'GSUB'
Took 0.000s to subset GlyphOrder
Took 0.000s to prune 'head'
Took 2.202s to prune 'CFF '
Took 0.000s to prune 'GDEF'
Took 0.002s to prune 'GPOS'
Took 0.005s to prune 'GSUB'
Took 0.186s to prune 'name'
Took 1.160s to compile and save font
Took 5.102s to make one with everything (TOTAL TIME)

real	0m5.222s
user	0m5.147s
sys	0m0.059s

Without name closure it takes 4.925 (as in without this PR). The name pruning took 0.186s, which I think is acceptable.

With the previous approach this was taking minutes before I interrupted it...

[imran-iq]

it would be nice if @imran-iq could give this PR branch a try with their go-noto font build to confirm it is fast enough?

Little late to the party, but running locally with these changes (just used main) it does not get stuck at the pyftsubset step like in #3213