[PATCH] Avoid deadlock in malloc on backtrace

[Paul Eggert]

Siddhesh Poyarekar wrote:
> A new function backtrace_init() could help break this deadlock.

Or we could document the current behavior, which is that backtrace (0, 0) 
initializes the backtrace module and avoids the problem.  This is pretty much 
what bleeding-edge Emacs does now.  (Unfortunately this does not work on sparc 
or microblaze but those would presumably be considered bugs.)