remote/performance: don't construct a merkle tree for remote caching

[buchgr]

Remote caching doesn't benefit from constructing a merkle. Some users reported significant performance overheads from constructing the merkle tree. We might just want to compute the hash of the list of input file hashes instead.

cc: @benjaminp

[buchgr]

building a merkle tree should not be much more expensive than building a list if implemented correctly.

[Globegitter]

Does this still hold true compared to the comment mentioned here: #7583 (comment)?

[buchgr]

I don't know. You'd have to benchmark the current merkle tree implementation against a list based approach I guess. I'd expect the list based approach to still be faster but I am not sure whether it's by a lot.