Specify how to disable timeouts

[dridi]

Following consensus, this patch series formalizes several things:

• NAN means that a timeout is not set (only possible in VCL, parameters must have a default value)
• it is possible to unset timeout variables in VCL
  • they are unset by default (it was already the case)
  • they fall back to the relevant parameter when read (it was already the case for sess.*timeout*)
• INFINITY disables a timeout
  • it translates to VRT_DECIMAL_MAX in VCL
  • in the poll(2) case it translates to -1 (aka INFTIM)
  • in the setsockopt(2) case it translates to a zero timeval
  • a zero timeout for setsockopt(2) translates to a 1us timeout (should it be 1ms?)
• a new duration tweak is added for parameters (inconsequential for designated parameters)
• the existing timeout tweak can now take a special value "never"
• bereq timeouts have two levels of fallback
  • backend timeouts defined in VGC can't use INFINITY (no math.h)
  • backend timeouts default to a negative value (the lack of zero override was also reported by @walid-git)
  • this is a VRT API breakage! (not the ABI)

I tried to be as thorough as possible, following the trail of relevant timeout parameters.

Fixes #3045

[dridi]

Of course I had to forget something, and it was Lck_CondWaitTimeout(), which I already added to the patch series. I decided to be prudent and keep support for zero meaning never timing out, but looking closely the only case where we use zero is actually in the Lck_CondWait() function, and Lck_CondWaitTimeout() when we forward a zero to Lck_CondWaitUntil().

So I think we could safely make zero no longer special, and in that case have a 1ms timed wait, or less.