Precompute nock hashes

[lukaszcz]

• Closes Precompute hashes of Nock terms and store them in Cells & Atoms #3383
• Depends on: Nock modularization #3370

Benchmarks

• juvix-old - the version from Nock modularization #3370
• juvix - the version from this PR

Both compiled with optimizations.

$ hyperfine -w 1 -p 'rm -rf .juvix-build' 'juvix-old compile anoma test001.juvix --modular' 'juvix compile anoma test001.juvix --modular'

Benchmark 1: juvix-old compile anoma test001.juvix --modular
  Time (mean ± σ):      6.178 s ±  0.041 s    [User: 12.155 s, System: 0.879 s]
  Range (min … max):    6.101 s …  6.238 s    10 runs
 
Benchmark 2: juvix compile anoma test001.juvix --modular
  Time (mean ± σ):      5.775 s ±  0.018 s    [User: 11.636 s, System: 0.900 s]
  Range (min … max):    5.750 s …  5.794 s    10 runs
 
Summary
  'juvix compile anoma test001.juvix --modular' ran
    1.07 ± 0.01 times faster than 'juvix-old compile anoma test001.juvix --modular'

$ hyperfine -w 1 'juvix-old compile anoma test001.juvix --modular' 'juvix compile anoma test001.juvix --modular'

Benchmark 1: juvix-old compile anoma test001.juvix --modular
  Time (mean ± σ):      2.160 s ±  0.049 s    [User: 3.540 s, System: 0.829 s]
  Range (min … max):    2.092 s …  2.234 s    10 runs
 
Benchmark 2: juvix compile anoma test001.juvix --modular
  Time (mean ± σ):      1.811 s ±  0.031 s    [User: 3.209 s, System: 0.847 s]
  Range (min … max):    1.769 s …  1.856 s    10 runs
 
Summary
  'juvix compile anoma test001.juvix --modular' ran
    1.19 ± 0.03 times faster than 'juvix-old compile anoma test001.juvix --modular'

$ hyperfine -w 1 'juvix-old dev nockma run builtin-evaluator test001.nockma --storage test001.modules.nockma' 'juvix dev nockma run builtin-evaluator test001.nockma --storage test001.modules.nockma'

Benchmark 1: juvix-old dev nockma run builtin-evaluator test001.nockma --storage test001.modules.nockma
  Time (mean ± σ):      6.693 s ±  0.056 s    [User: 7.388 s, System: 1.212 s]
  Range (min … max):    6.614 s …  6.811 s    10 runs
 
Benchmark 2: juvix dev nockma run builtin-evaluator test001.nockma --storage test001.modules.nockma
  Time (mean ± σ):      6.933 s ±  0.060 s    [User: 7.650 s, System: 1.321 s]
  Range (min … max):    6.783 s …  6.997 s    10 runs
 
Summary
  'juvix-old dev nockma run builtin-evaluator test001.nockma --storage test001.modules.nockma' ran
    1.04 ± 0.01 times faster than 'juvix dev nockma run builtin-evaluator test001.nockma --storage test001.modules.nockma'

Conclusion

• Precomputing the hashes increases the performance of jamming, which dominates the final stage of compilation to Nock.
• The additional field in the CellInfo datatype decreases evaluation performance, which is not compensated by the faster module load times. However, compilation is the main user-facing feature of the Juvix compiler.

[janmasrovira]

why not put _cellHash inside _cellInfo?

[lukaszcz]

yes, we could do that. I'll change it