Add support for epoch callbacks

[bcumming]

Adds support for the user to add a generic callback that is to be called:

1. before the first time step
2. at the end of every epoch

One implementation of such a callback that prints a progress bar to the screen is provided, and wrapped for the Python front end. Example output of the progress bar:

 57% [||||||||||||||||||||||||||||                      ]           575ms

Fixes #1870

[jlubo]

The callback and the Python wrapper seem to work well, and the progress bar looks fine!

See below for two minor suggestions. Also, documentation is still missing.

[jlubo]

Suggested change

	fflush(stdout);
	if (t == tfinal) {
	printf("\n");
	}
	fflush(stdout);

Automatically adding a linebreak after the simulation has finished.

[bcumming]

👍

I tweaked it to print a new line and reset the counter so that code like this:

sim.run(1000, 0.1)
sim.run(2000, 0.1)
sim.run(3000, 0.1)

will generate the following output:

100% |--------------------------------------------------|          1000ms
100% |--------------------------------------------------|          2000ms
100% |--------------------------------------------------|          3000ms

[jlubo]

Suggested change

	print('\nSimulation finished')
	print('Simulation finished')

Linebreak should have been added automatically (see above).

[jlubo]

Suggested change

	const char* PBSTR = "||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||";
	unsigned PBWIDTH = 50;
	unsigned PBWIDTH = 50;
	const std::string PBSTR = std::string(2*PBWIDTH, '|');

Using the "repeat character" constructor of std::string.

[bcumming]

The benefit of const char* is that the buffer doesn't have to be initialized each time the callback is made, whereas a std::string requires memory allocation on every callback (at least until we can have constexpr std::string).
Instead I used static const std::string that will only be initialized once.

[jlubo]

Seems like the best solution!

[jlubo]

Suggested change

	printf("\r%3d%% [%.*s%*s] %12ums", val, lpad, PBSTR, rpad, "", (unsigned)t);
	printf("\r%3d%% [%.*s%*s] %12ums", val, lpad, PBSTR.c_str(), rpad, "", (unsigned)t);

I had forgot to mention this: the c_str() conversion also has to be used if the std::string is used (see above).

[thorstenhater]

One tiny comment, rest is good.

[thorstenhater]

global is redundant? Or is this compatibility with spike_callback?

[bcumming]

Good point. That is a hang over from an earlier draft of the epoch callback.
I realised that the global information is redundant, and the caller can decide whether to attach a callback on one or more distributed contexts for themselves, e.g.:
https://github.com/arbor-sim/arbor/pull/1873/files#diff-5827156d5c4d2c924b0def9bbf4e51db85e1ea4a0c45c75328ca38dab6ea8e12R190

[thorstenhater]

👍