Trajectory realloc

[molpopgen]

Skips the really big allocation of currentTrajectory in favor of realloc.

Changes:

• currentTrajectory is no longer global. This was not required, but it makes it more obvious that the variable is owned by main.
• proposeTrajectory now takes a pointer to currentTrajectory, e.g. float **. This change is required so that realloc events are propagated back to the caller. The more idiomatic approach would be to have this function return the new pointer, but it already has a return value.
• MAXTRAJSIZE is a global size_t, defined as extern in discoal.h and initialized in discoal_multipop.c. This variable is used to track the allocation size.
• Error handling is done via exit rather than assert, which would normally be disabled.

Note that MAXTRAJSIZE is set too small. This was intentional, in order to stress test the realloc code.

I also thing I added -g to the Makefile as I was tracking down a double free issue for a bit there...

[andrewkern]

i've been dragging my heels on this PR as it changes a lot of code. Have you been using this successfully? no issues?

[molpopgen]

My inclination would be to not merge this until a formal code review is conducted and/or unit tests are put in place.

I did use this for the case of running a single replicate, and results were identical to the current release. A good test would be to run a lot of replicates with both versions and compare output, but that wouldn't make a good integrated test.

The other oddity is that the VIRT allocations are still massive. Not sure why. When I get time, I'll throw some stronger profilers at the code, like a full-blown valgrind session, etc.

[molpopgen]

Closing due to inactivity.