[BFS] Optimize BlockAllocator::AllocateBlocks' block-search routine

[waddlesplash]

At present, it just uses the fFirstFree to find the first free block and then (if the other sanity checks succeed) looks for free blocks one bit at a time. Instead, it could use bit-operations like fls as well as just checking if all the bits in an int32 are set, which should significantly speed things up.

On partitions with fragmentation, BlockAllocator::AllocateBlocks can form a large part of the total time needed to write files (in one profiling trace X512 sent to me, it was over 80% of the total; though admittedly the profiler could be mistaken.)