LocalSet optimisations

[aumetra]

Optimise some parts of the code by making use of the fact that a LocalSet can execute !Send futures.

This is mostly an experiment, seeing whether the additional allocations to simply being able to send the values to the different executor thread are even worth it.

[aumetra]

An example for the target of the optimisations is the parsing of the mentions. The parsing itself is very quick, microseconds even, but we need a Mutex to synchronize accesses to the accumulator.

That synchronization overhead could be avoided using a LocalSet even though the atomics overhead should be negligible.

Another thing this would allow us is also to introduce bump allocation to certain paths in the code, since allocation arenas are inherently !Send.

Again, whether the overhead of the triple boxing+channel send is worth it in the end has to be checked.