Pre-Allocating Frozen Tier ShardBytes Can be very Slow on Windows

[original-brownbear]

There are two issues with pre-allocating the frozen tier shared bytes on windows:

1. On Windows we do not use a FS native way like fallocate on Linux to pre-allocate the large shared cache file. This can take a significant amount of time for larger files on slower disks. We should look into a native way to speed the process up and can probably also speed it up directly by increasing the write chunks size we use when allocating the file form the current 8k.
2. On node restart we will try to pre-allocate the file and fail on the attempt to do it via a native call. As a result we delete only to then write it again from scratch, thus running into the first problem again on node restarts.

[elasticmachine]

Pinging @elastic/es-distributed (Team:Distributed)