Derive other indexes directly for binary fuse

[RaduBerinde]

We manipulate the math and use bit tricks to derive the other two
indexes more efficiently during peeling.

Apple M1:

name                                old MKeys/s    new MKeys/s    delta
BinaryFusePopulate/8/n=10000-10         43.8 ± 2%      50.3 ± 3%  +14.88%  (p=0.000 n=8+9)
BinaryFusePopulate/8/n=100000-10        38.6 ± 3%      41.3 ± 1%   +7.09%  (p=0.000 n=9+8)
BinaryFusePopulate/8/n=1000000-10       35.0 ± 4%      36.5 ± 7%   +4.12%  (p=0.013 n=9+10)
BinaryFusePopulate/16/n=10000-10        48.6 ± 4%      48.5 ± 6%     ~     (p=1.000 n=10+10)
BinaryFusePopulate/16/n=100000-10       38.0 ± 3%      41.1 ± 1%   +8.35%  (p=0.000 n=10+10)
BinaryFusePopulate/16/n=1000000-10      33.8 ± 5%      36.6 ± 2%   +8.14%  (p=0.000 n=10+10)

GCE N4D (AMD Turin):

name                               old MKeys/s    new MKeys/s    delta
BinaryFusePopulate/8/n=10000-8         53.2 ± 3%      57.1 ± 1%   +7.46%  (p=0.000 n=10+10)
BinaryFusePopulate/8/n=100000-8        33.0 ± 0%      37.5 ± 1%  +13.38%  (p=0.000 n=10+10)
BinaryFusePopulate/8/n=1000000-8       28.5 ± 2%      31.8 ± 2%  +11.59%  (p=0.000 n=10+10)
BinaryFusePopulate/16/n=10000-8        53.1 ± 1%      56.2 ± 1%   +5.93%  (p=0.000 n=10+10)
BinaryFusePopulate/16/n=100000-8       31.8 ± 1%      37.3 ± 1%  +17.35%  (p=0.000 n=10+10)
BinaryFusePopulate/16/n=1000000-8      27.5 ± 1%      30.9 ± 1%  +12.34%  (p=0.000 n=10+10)

GCE C4 (Intel Emerald Rapids, turbo boost capped at "all core" max):

name                               old MKeys/s    new MKeys/s    delta
BinaryFusePopulate/8/n=10000-8         29.2 ± 1%      32.2 ± 1%  +10.00%  (p=0.000 n=10+10)
BinaryFusePopulate/8/n=100000-8        27.0 ± 3%      29.8 ± 5%  +10.22%  (p=0.000 n=10+10)
BinaryFusePopulate/8/n=1000000-8       25.6 ± 3%      28.2 ± 5%  +10.27%  (p=0.000 n=10+10)
BinaryFusePopulate/16/n=10000-8        28.9 ± 1%      32.0 ± 1%  +10.84%  (p=0.000 n=10+10)
BinaryFusePopulate/16/n=100000-8       26.2 ± 1%      28.8 ± 3%  +10.05%  (p=0.000 n=10+10)
BinaryFusePopulate/16/n=1000000-8      24.8 ± 2%      26.9 ± 2%   +8.37%  (p=0.000 n=10+10)

GCE C4A (Google's Axion ARM64):

name                               old MKeys/s    new MKeys/s    delta
BinaryFusePopulate/8/n=10000-8         45.1 ± 1%      45.1 ± 1%    ~     (p=0.511 n=9+10)
BinaryFusePopulate/8/n=100000-8        39.8 ± 1%      39.4 ± 1%  -0.79%  (p=0.018 n=9+10)
BinaryFusePopulate/8/n=1000000-8       33.9 ± 3%      34.2 ± 3%    ~     (p=0.363 n=10+10)
BinaryFusePopulate/16/n=10000-8        44.0 ± 1%      44.7 ± 1%  +1.54%  (p=0.000 n=9+10)
BinaryFusePopulate/16/n=100000-8       37.4 ± 1%      38.4 ± 1%  +2.75%  (p=0.000 n=10+10)
BinaryFusePopulate/16/n=1000000-8      30.9 ± 5%      32.4 ± 1%  +4.84%  (p=0.000 n=10+10)

[lemire]

On my todo to review.