In this PR, we make the planner take the scheduling decisions in a centralized manner. Intuitively, this means that the callFunctions method is moved from the Scheduler to the PlannerClient.

This PR finalizes most of the work to move from a decentralised to a centralized scheduler, and it includes some relevant changes in functionality (for the tests mostly):

• By default, we don't overload the executor pool. This means that we need to make sure that the planner sees that the nodes have enough cores to run the requests (even in GHA).
• Each execution request involves a request round-trip to the planner (a background thread in the tests), which means that we need to be more careful to adequately wait for results (particularly for sanitized builds).
• Mocking multiple nodes (e.g. for remote threads) becomes harder, as we can't mock one but not the other one. So all remote thread's tests need to happen in the distributed tests, or using a lower-level interface (i.e. not callFunctions but executeBatch or executeThreads).

This PR also removes a lot of functionality from the Scheduler class. This class had become very overloaded, and now is almost unnecessary. In subsequent PRs I will separate the Executor from the Scheduler and decide exactly how to rename the remainders of the Scheduler.

Closes #275
Closes #287