ghost: support marathons

[matthewearl]

Fixes #63

Still needs a bit more testing, but I've tried it out with Muty's 12:11 id1. On my machine loading end with the ghost enabled is apparently instant, which should be the slowest map to load since it has to go through the entire demo.

[matthewearl]

I've added a new commit which adds support for multi-demo marathons chained with a stuff_text "playdemo". Updated the PR title to reflect this.

[matthewearl]

There's a merge conflict on this branch with #60 relating to whether Ghost_ReadDemo takes a char * or a FILE *, I'll sort it out when the time comes.

[matthewearl]

I've added a couple of commits which track ghost split times for marathons

The tricky bit in this commit is informing the client when the marathon starts. I've handled this by adding a command marathon which is stufftext'd by the server after sending the server info indicating whether this is the first, or the subsequent level of a marathon.

[matthewearl]

I've fixed the conflicts. Separate to that, there's a slight annoyance with the marathon command in that when you playback a demo recorded with this version of joequake on an older version (or not joequake at all) then you get an "unknown command: marathon" message at the start of each level. I could instead hack some other command to hide it entirely? For example, I could encode it in the looptrack argument of svc_cdtrack which as far as I can tell has always been ignored. Thoughts?

[matthewearl]

In fact, if I use an svc_print in the server info packet it doesn't appear on the screen (it appears when you pull down the console). I should be able to eliminate the message on marathon start too, just requiring a message for marathon continuation. So this would just be a svc_print with a message like "Continuing JoeQuake marathon", which would be less alarming than an "unknown command" error on the screen on every map load, which can still be parsed programmatically. Also less hacky than the svc_cdtrack solution.

[j0zzz]

Yeah, I like the solution using the svc_print server command. Putting into the svc_cdtrack seemed like a nasty hack to me either.