Compute true_count while constructing/evaluating boolean operators

[gatesn]

We almost always want to know true_count from the result of mask operations, so we may as well compute them while iterating over the u64s used within BooleanBuffer.

We probably need our own version of BooleanBuffer::collect_bool that does a popcnt for each u64.

[connortsui20]

Now that we have our own BitBuffer type (vortex-buffer/src/bit/buf.rs), we should be able to lazily compute this and put the value in a OnceLock similar to how MaskValues (in vortex-mask/src/lib.rs) does this for its slices and/or indices.

[joseph-isaacs]

I think we want to compute this along with packing the bits? No once lock.

[robert3005]

For all the cases we have tried optimised double pass is faster than fused single pass. I imagine there's some size threshold when we are no longer cache resident and single pass wins but for most practical cases this number is far larger than most individual array sizes we create.