Force Merge should check disk space before attempting a merge to a single segment

[markharwood]

We know that a force merge down to a single segment will be roughly a 2x cost of the current index size because it requires the old and new index co-exist on disk (briefly) before swapping over to the new index.
We should introduce a disk space check before attempting such a merge and return an error if the operation is estimated to tip us over the high watermark threshold for allocating shards to the current node.
This would save the heavy cost of attempting an impossible task and prevent the dangerous scenario of running out of disk space.

[elasticmachine]

Pinging @elastic/es-core-infra

[geekpete]

It's probably good to check continually through the force merge process if possible? Rather than only at the start, in case things change while it's merging after it's done the initial calculation. This is probably only useful if there's a mechanism to pause or cancel the force merge if the disk space check fails part way through, I'm unsure if internally we have any existing way to pause/cancel a force merge.