Remove tag from RuntimeValue

[eira-fransham]

This closes #99. Benchcmp results:

 name               preopt.bench ns/iter  postopt.bench ns/iter  diff ns/iter   diff %  speedup 
 bench_regex_redux  4,471,532             4,117,821                  -353,711   -7.91%   x 1.09 
 bench_rev_comp     10,363,517            9,552,537                  -810,980   -7.83%   x 1.08 
 bench_tiny_keccak  5,475,809             5,044,934                  -430,875   -7.87%   x 1.09 
 fac_opt            42,472                26,942                      -15,530  -36.57%   x 1.58 
 fac_recursive      43,852                28,770                      -15,082  -34.39%   x 1.52 
 recursive_ok       1,423,602             1,363,951                   -59,651   -4.19%   x 1.04 
 recursive_trap     182,536               160,872                     -21,664  -11.87%   x 1.13 

All tests still pass. I expect most of these improvements are due to the fact that we double our cache efficiency by halving the size of RuntimeValue (since RuntimeValue includes a u64, its tag must be padded to 64 bits, making the whole value 2x the size of a u64). However, some improvements may be due to the fact that we remove some validity checking on function arguments. This was duplicated effort anyway, since that is already checked statically.

[pepyakin]

Can you add a brief comment what is going on here and why?

[eira-fransham]

Great point, this is definitely not self-explanatory

[pepyakin]

Could you rebase this please?

[eira-fransham]

Per mine and @pepyakin's discussion in private, we seem to get such a huge improvement on the factorial benchmarks because zeroing allocation can use memset/calloc now instead of having to write bytes individually. Before it was setting stack values to I32(0), which (evidently) doesn't have the tag 0 and so memset(0) wouldn't work, but now because it's essentially just writing a zeroed u64 Rust can generate a memset or calloc (depending on whether we're on the stack or the heap, respectively).

[pepyakin]

Could you rebase this?

[eira-fransham]

@pepyakin After this rebase can you just squash+merge (unless it's incorrect of course) because I have to manually merge every time because all our changes touch runner.rs

[pepyakin]

yeah sure! Just ping me in every possible way you can whenever you've done with rebasing : )

[pepyakin]

Perfect, thank you!