Stats now prints cache hits

[gabriel-barrett]

Example:

=== Circuit Statistics ===
Circuits: 626
Total width: 48210
Total FFT cost: 348748181
Total cache hits: 336151
Total saved cost: 32.65%
------------------------------------------------------------------------------------------------------------------
Name                                                                   Width Height  Hits FFT cost       %     %++
------------------------------------------------------------------------------------------------------------------
u32_add                                                                   60  87952   332 86673682  24.85%  24.85%
blake3_g_function                                                        210  14714    14 42779916  12.27%  37.12%
u32_xor                                                                   38  60914    46 36791476  10.55%  47.67%
blake3_compress_chunks                                                    75  14692     0 15253319   4.37%  52.04%
k_infer                                                                  188   6031  1207 14238818   4.08%  56.13%
expr_inst1                                                                60  16108  7161 13507031   3.87%  60.00%
whnf_with_spine                                                          179   5437   180 12076330   3.46%  63.46%
expr_inst1_walk                                                           73  11520     0 11346109   3.25%  66.71%
memory[3]                                                                 14  35123 82952  7425065   2.13%  68.84%
list_lookup_u64.Ptr.Univ                                                  49  10530    69  6894504   1.98%  70.82%
...

Note that the saved cost is a bit misleading. In general, the real saved cost is much more, because it is not considering recursive calls. For instance, if you call a function factorial 4 twice, the cache hit will not only save the need to prove factorial 4 again, but also factorial 3, factorial 2, etc