Add support for asynchronous workers

[nefrathenrici]

Purpose

This PR enables workers to be added while a calibration is already underway.

To-do

• Add redirect_stderr(stdout) to worker logger func

Content

• Add an expr field to PBSManager and SlurmManager. This expr is evaluated just after the worker is initialized. This can be set via SlurmManager(nprocs; expr). This enables workers to be initialized and load code asynchronously, without relying on @everywhere.
• Add method specialization for Distributed.create_worker(manager::Union{SlurmManager, PBSManager},...) in order to ensure the expression gets evaluated before any other code.
• Changed run_worker_iteration to check for newly added workers for each forward model run. The function now gets initial workers, then makes a vector of work (fwd model calls). While this vector is not empty, we check for new workers, and if workers are available we assign one unit of work to the worker. If no workers are available, wait for workers

Previous workflow:

addprocs(SlurmManager(nprocs)...)
@everywhere
    include(model_interface_file)
end

calibrate(WorkerBackend, ...)

New workflow:

expr = quote
    include(model_interface_file)
end
@async addprocs(SlurmManager(nprocs; expr)...)

calibrate(WorkerBackend, ...)

---

• I have read and checked the items on the review checklist.

[nefrathenrici]

This is incorrect and should be redirect_stderr(io)