Use FastHash on Piece.

[munificent]

This mixin overrides the default identity-based hash code with a stored integer hash code. That made a big difference in the old formatter's performance but for most of the times I tried it with the new one, it didn't help. But it seems that the new formatter is storing pieces in maps enough for it to make a difference.

On the microbenchmarks:

Benchmark (tall)                fastest   median  slowest  average  baseline
-----------------------------  --------  -------  -------  -------  --------
block                             0.057    0.061    0.121    0.065    107.0%
chain                             0.524    0.539    0.579    0.541    117.1%
collection                        0.247    0.256    0.273    0.257    107.1%
collection_large                  1.539    1.559    1.621    1.562    106.0%
conditional                       0.060    0.061    0.079    0.062    107.0%
curry                             0.455    0.462    0.477    0.463    118.0%
ffi                               0.143    0.149    0.159    0.148    105.4%
flutter_popup_menu_test           0.503    0.514    0.541    0.516    115.0%
flutter_scrollbar_test            0.399    0.406    0.450    0.410    113.2%
function_call                     1.658    1.709    1.849    1.716    106.0%
infix_large                       0.552    0.567    0.600    0.570    108.9%
infix_small                       0.145    0.148    0.162    0.149    105.3%
interpolation                     0.087    0.089    0.101    0.090    102.5%
interpolation_1516                0.070    0.071    0.085    0.071    100.7%
large                             3.263    3.280    3.573    3.298    109.0%
top_level                         0.127    0.128    0.147    0.130    107.9%

And when formatting the Flutter repo:

Current formatter    10.138 ========================================
Optimized             9.490 =====================================
Old formatter         4.812 ==================
The current formatter is 52.54% slower than the old formatter.
The optimized is 6.83% faster than the current formatter.
The optimized is 49.29% slower than the old formatter.
The optimization gets the formatter 12.17% of the way to the old one.

That's pretty good for a one line change.

Also, this makes debugging the new formatter easier because now each Piece has a different debug string. Trying to figure out which of a hundred Infix pieces is the one you want gets pretty tedious...