Recursive critical sections

[Nicolas]

If an application uses critical sections recursively, or uses a critical section during a checkpoint, the first boinc_end_critical_section call ends the critical section "prematurely". It would be better to change in_critical_section to be an int instead of a bool to keep the recursion count, and consider it shouldn't suspend if in_critical_section > 0

I'm attaching a patch that does this, to lower the work needed in case you actually think my enhancement suggestion is a good idea :) I changed !in_critical_section to >0. This is technically not needed since (!foo) == (foo!=0), but I think it's cleaner to have the comparison explicit.

_{(of course, feel free to resolve as wontfix if you disagree on the usefulness of the change)}

[Nicolas]

Patch

[Nicolas]

Modified patch, abstracting crit. sections from checkpoint code.

[Nicolas]

Wait, that's totally wrong. !in_critical_section should be replaced with in_critical_section == 0, not in_critical_section > 0!

[Nicolas]

Fixed stupid mistake (inverted conditions)

[davea]

(In [14694]) - API: critical section begin/end calls can be nested

Fixes #508 (from Nicolas)