feat!: high precision random number generator

[ST-DDT]

Fixes #2355

• faker.datatype.number often gives duplicates #2355

To Fix

• fix/reduce duplicates returned from faker.number.int() and faker.number.float()
• fix faker.number.int() only returning even numbers (31bits out of 53)
• fix faker.number.int(max) is incapable of returning max value for large maxes

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 99.57%. Comparing base (0d4cba6) to head (a4b4203).

❗ Current head a4b4203 differs from pull request most recent head 998611b. Consider uploading reports for the commit 998611b to get more accurate results

Additional details and impacted files

@@            Coverage Diff             @@
##             next    #2357      +/-   ##
==========================================
- Coverage   99.57%   99.57%   -0.01%     
==========================================
  Files        2846     2815      -31     
  Lines      249892   249870      -22     
  Branches     1055     1066      +11     
==========================================
- Hits       248830   248804      -26     
- Misses       1062     1066       +4     

Files	Coverage Δ
src/faker.ts	100.00% <100.00%> (ø)
src/index.ts	100.00% <100.00%> (ø)
src/internal/mersenne.ts	97.58% <100.00%> (+0.14%)	⬆️
src/simple-faker.ts	100.00% <100.00%> (ø)

... and 83 files with indirect coverage changes

[matthewmayer]

Probably also worth checking if genrandRes53CC is significantly slower, or if there are any other downsides to using it.

[ST-DDT]

It's twice as slow.

[matthewmayer]

I think this should be probably considered a breaking change given it causes nearly every snapshot to change. That would also allow for beta testing of the change before 9.0 in case the slower algorithm causes performance problems.

[ST-DDT]

The breaking of the snapshots is caused by next() now effectively taking 2 values from the seed.
We could reduce this if we dynamically request either 32/53 bits of precision depending on the values in use.
e.g. an array with length 15 doesn't need 53 bits of precision for the index.

[ST-DDT]

I'm not that deep into math, but for some reason using two values from the seed increases the variance of the values.
Which increase likelihood of this test to fail: https://github.com/faker-js/faker/actions/runs/6025687970/job/16346941314?pr=2357

[ST-DDT]

I'm not that deep into math, but for some reason using two values from the seed increases the variance of the values. Which increase likelihood of this test to fail: faker-js/faker/actions/runs/6025687970/job/16346941314?pr=2357

Nevermind, I found the error, currently the first and the last element aren't evenly distributed for number.int()

[ST-DDT]

Team Decision

We will change mersenne to 53 bit precision to reduce the duplicates and create a separate PR for the float issues.

[ST-DDT]

3 approving reviews. Enabling auto merge.