Sprinkle yk_is_interpreting as appropriate.

[ltratt]

Needs ykjit/yk#2113.

This speeds up a number of benchmarks. On b16 I get:

 LuLPeg/lua/              11861 ±  79   10287 ± 174   0.87  13.28% faster
 json/lua/100             10952 ± 133   10228 ± 242   0.93  6.61% faster
 storage/lua/1000         22579 ± 156   21776 ± 272   0.96  3.55% faster
 spectralnorm/lua/1000     9132 ±  43    8808 ±  64   0.96  3.55% faster
 havlak/lua/1500          73547 ± 390   70956 ± 223   0.96  3.52% faster
 cd/lua/250               31251 ± 280   30296 ± 265   0.97  3.05% faster
 HashIds/lua/6000          8825 ± 153    8557 ±  97   0.97  3.04% faster

b15 is a bit less impressive, but still there are 5 benchmarks in the 3-5% improvement range.

[ltratt]

This won't work until the yk PR is merged.