Use new rand_core::block::Generator trait

[dhardy]

• Added a CHANGELOG.md entry

Summary

Use rust-random/rand_core#26.

Details

This worked out nicely:

• No need for Array64 since we can (and must!) use [u32; 64]
• Benchmarks show no significant effect on rand::rng():

random_bytes/thread     time:   [307.01 ns 307.25 ns 307.48 ns]
                        thrpt:  [3.1016 GiB/s 3.1039 GiB/s 3.1064 GiB/s]
                 change:
                        time:   [-4.5564% -4.4324% -4.3065%] (p = 0.00 < 0.05)
                        thrpt:  [+4.5003% +4.6379% +4.7739%]
                        Performance has improved.
Found 24 outliers among 100 measurements (24.00%)
  9 (9.00%) low severe
  13 (13.00%) high mild
  2 (2.00%) high severe

random_u32/thread       time:   [1.2800 ns 1.2801 ns 1.2801 ns]
                        thrpt:  [2.9101 GiB/s 2.9102 GiB/s 2.9104 GiB/s]
                 change:
                        time:   [+0.3641% +0.4067% +0.4495%] (p = 0.00 < 0.05)
                        thrpt:  [-0.4475% -0.4050% -0.3628%]
                        Change within noise threshold.
Found 145 outliers among 1000 measurements (14.50%)
  25 (2.50%) low severe
  16 (1.60%) low mild
  27 (2.70%) high mild
  77 (7.70%) high severe

random_u64/thread       time:   [2.0227 ns 2.0235 ns 2.0243 ns]
                        thrpt:  [3.6806 GiB/s 3.6821 GiB/s 3.6835 GiB/s]
                 change:
                        time:   [-0.8368% -0.7952% -0.7544%] (p = 0.00 < 0.05)
                        thrpt:  [+0.7601% +0.8016% +0.8439%]
                        Change within noise threshold.
Found 2 outliers among 1000 measurements (0.20%)
  1 (0.10%) high mild
  1 (0.10%) high severe

(For whatever reason the first bench in a run is often slightly different; others are very close to prior performance which is not the case in #1686.)

Draft because: depends on unmerged PR.