Add bit transpose operations

[robert3005]

Logic to convert bit buffers into transpose layout. This is useful where
intermediary arrays are in transpose layout

In particular this is necessary to handle DeltaArray validity correctly

On Zen 5 the bmi instructions are 20-30% faster, VBMI are ~10-20x faster
On Zen3 bmi instructions are still around 20-30% faster
On M4 the neon version is ~60-100% faster while untranspose stays the same

Full benchmark results for posterity
on Zen 5 machine (m8azn)

Compiling fastlanes v0.5.0 (/home/ubuntu/fastlanes)
    Finished `bench` profile [optimized] target(s) in 0.56s
     Running benches/bit_transpose.rs (target/release/deps/bit_transpose-dd4b19170a0386ff)
Timer precision: 10 ns
bit_transpose                      fastest       │ slowest       │ median        │ mean          │ samples │ iters
├─ transpose_scalar                59.8 ns       │ 1.059 µs      │ 69.8 ns       │ 75.7 ns       │ 100     │ 100
├─ transpose_scalar_throughput     8.349 µs      │ 15.75 µs      │ 8.369 µs      │ 8.443 µs      │ 100     │ 100
├─ untranspose_scalar              39.8 ns       │ 40.73 ns      │ 40.11 ns      │ 39.99 ns      │ 100     │ 3200
├─ untranspose_scalar_throughput   8.889 µs      │ 14.26 µs      │ 8.909 µs      │ 8.985 µs      │ 100     │ 100
╰─ x86                                           │               │               │               │         │
   ├─ transpose_bmi2               29.64 ns      │ 30.26 ns      │ 29.95 ns      │ 29.94 ns      │ 100     │ 6400
   ├─ transpose_bmi2_throughput    6.309 µs      │ 10.92 µs      │ 6.339 µs      │ 6.386 µs      │ 100     │ 100
   ├─ transpose_vbmi               3.608 ns      │ 3.667 ns      │ 3.628 ns      │ 3.631 ns      │ 100     │ 51200
   ├─ transpose_vbmi_throughput    504.8 ns      │ 524.8 ns      │ 509.8 ns      │ 508 ns        │ 100     │ 200
   ├─ untranspose_bmi2             27.61 ns      │ 70.42 ns      │ 27.76 ns      │ 28.24 ns      │ 100     │ 6400
   ├─ untranspose_bmi2_throughput  6.199 µs      │ 10.57 µs      │ 6.219 µs      │ 6.262 µs      │ 100     │ 100
   ├─ untranspose_vbmi             3.589 ns      │ 3.628 ns      │ 3.608 ns      │ 3.604 ns      │ 100     │ 51200
   ╰─ untranspose_vbmi_throughput  489.8 ns      │ 504.8 ns      │ 494.8 ns      │ 496.7 ns      │ 100     │ 200

on a Zen 3 machine (c6a)

Compiling fastlanes v0.5.0 (/home/ubuntu/fastlanes)
    Finished `bench` profile [optimized] target(s) in 1.02s
     Running benches/bit_transpose.rs (target/release/deps/bit_transpose-dd4b19170a0386ff)
Timer precision: 20 ns
bit_transpose                      fastest       │ slowest       │ median        │ mean          │ samples │ iters
├─ transpose_scalar                70.65 ns      │ 962.9 ns      │ 70.97 ns      │ 81.99 ns      │ 100     │ 3200
├─ transpose_scalar_throughput     15.68 µs      │ 37.39 µs      │ 15.72 µs      │ 16.04 µs      │ 100     │ 100
├─ untranspose_scalar              72.84 ns      │ 242.2 ns      │ 75.04 ns      │ 77.92 ns      │ 100     │ 3200
├─ untranspose_scalar_throughput   16.13 µs      │ 32.4 µs       │ 16.26 µs      │ 16.69 µs      │ 100     │ 100
╰─ x86                                           │               │               │               │         │
   ├─ transpose_bmi2               58.75 ns      │ 229.4 ns      │ 59.09 ns      │ 61.18 ns      │ 100     │ 3200
   ├─ transpose_bmi2_throughput    12.33 µs      │ 27.84 µs      │ 12.41 µs      │ 12.68 µs      │ 100     │ 100
   ├─ transpose_vbmi               warning: No benchmark function registered for 'transpose_vbmi'

   ├─ transpose_vbmi_throughput    warning: No benchmark function registered for 'transpose_vbmi_throughput'

   ├─ untranspose_bmi2             57.22 ns      │ 452.2 ns      │ 58.78 ns      │ 63.58 ns      │ 100     │ 3200
   ├─ untranspose_bmi2_throughput  13.23 µs      │ 25.47 µs      │ 13.29 µs      │ 13.61 µs      │ 100     │ 100
   ├─ untranspose_vbmi             warning: No benchmark function registered for 'untranspose_vbmi'

   ╰─ untranspose_vbmi_throughput  warning: No benchmark function registered for 'untranspose_vbmi_throughput'

on a M4 Max

Finished [`bench` profile [optimized]](https://doc.rust-lang.org/cargo/reference/profiles.html#default-profiles) target(s) in 0.02s
     Running benches/bit_transpose.rs (target/release/deps/bit_transpose-9780c0103c6d2dc3)
Timer precision: 41 ns
bit_transpose                      fastest       │ slowest       │ median        │ mean          │ samples │ iters
├─ transpose_scalar                24.48 ns      │ 25.78 ns      │ 24.97 ns      │ 25.03 ns      │ 100     │ 25600
├─ transpose_scalar_throughput     5.249 µs      │ 9.166 µs      │ 5.332 µs      │ 5.545 µs      │ 100     │ 100
├─ untranspose_scalar              25.79 ns      │ 37.5 ns       │ 26.44 ns      │ 28.07 ns      │ 100     │ 12800
├─ untranspose_scalar_throughput   5.457 µs      │ 7.999 µs      │ 5.583 µs      │ 5.821 µs      │ 100     │ 100
╰─ aarch64                                       │               │               │               │         │
   ├─ transpose_neon               14.23 ns      │ 20.57 ns      │ 14.55 ns      │ 15.22 ns      │ 100     │ 25600
   ├─ transpose_neon_throughput    3.187 µs      │ 4.52 µs       │ 3.228 µs      │ 3.262 µs      │ 100     │ 200
   ├─ untranspose_neon             23.67 ns      │ 35.06 ns      │ 24.32 ns      │ 25.19 ns      │ 100     │ 25600
   ╰─ untranspose_neon_throughput  4.874 µs      │ 7.082 µs      │ 4.958 µs      │ 5.174 µs      │ 100     │ 100Finished 

Signed-off-by: Robert Kruszewski github@robertk.io

[joseph-isaacs]

LG otherwise