[ttx_diff] Add --timings flag, print fn runtimes

[cmyr]

speculative, we mentioned this a while ago and I got nerd-sniped.

for the example included below, we're spending lots of time in name_id_to_name. In rubik, we're spending lots of time normalizing off-by-ones.

---

This makes it easier to diagnose where exactly ttx_diff itself is spending time. Output looks like this:

TIMINGS
  0.658s  build fontc
  35.915s  build fontmake
  17.134s  ttx fontc
  17.053s  ttx fontmake
  0.443s  normalize fontc gpos
  0.314s  normalize fontmake gpos
  0.004s  normalize fontc gdef
  0.003s  normalize fontmake gdef
  1.142s  fill_in_gvar_deltas
  53.099s  reduce_diff_noise
    1.665s  sort indices
    13.976s  name id to name
    1.596s  gdef work
    14.911s  name id to name
    1.024s  gdef work
    3.007s  normalize glyf contours
    3.556s  normalize gvar contours
    7.455s  allow off-by-ones
  5.949s  extract_comparables fontc
  4.342s  extract_comparables fontmake
  0.332s  check_sizes
  2m29.088s  total

[wmedrano]

speculative, we mentioned this a while ago and I got nerd-sniped.

I'm so sorry Colin

lgtm, some lint errors need fixing on the Python side

[cmyr]

speculative, we mentioned this a while ago and I got nerd-sniped.

I'm so sorry Colin

lgtm, some lint errors need fixing on the Python side

No apologies necessary! I wouldn't have expected the off-by-ones to be such a slow operation, neat to be able to see that clearly.

[rsheeter]

In rubik, we're spending lots of time normalizing off-by-ones.

Worth citing #1932 which observes "Most of the time seems to be in ttx_diff in allow_some_off_by_ones"

[rsheeter]

(I'm pretty sure I wrote allow_some_off_by_ones off the cuff with a focus on v.simple outlines so it's not surprising there are fonts that it handles poorly)