Temporary fix for the Julia GC.

[rbehrends]

This addresses a problem where the Julia GC during a partial sweep frees
some, but not all objects of an unreachable data structure. If an
address to a remaining object of such a data structure is still randomly
hit during conservative stack scanning, it may erroneously try to mark
the deallocated objects, too.

This temporary fix works by validating each pointer before it is marked.
Because this is potentially expensive, the eventual solution should fix
the root cause, which is conservative stack scanning resurrecting
pointers to dead objects that have been freed during a partial
collection.

[coveralls]

Coverage decreased (-0.007%) to 85.222% when pulling 49f65b4 on rbehrends:julia-gc-validate-strict into 9886081 on gap-system:master.

[codecov]

Codecov Report

Merging #3412 into master will increase coverage by 0.01%.
The diff coverage is 66.66%.

@@            Coverage Diff             @@
##           master    #3412      +/-   ##
==========================================
+ Coverage    85.4%   85.41%   +0.01%     
==========================================
  Files         698      697       -1     
  Lines      344337   342390    -1947     
==========================================
- Hits       294097   292469    -1628     
+ Misses      50240    49921     -319

Impacted Files	Coverage Δ
src/weakptr.c	97.89% <100%> (-0.03%)	⬇️
src/julia_gc.c	77.4% <57.69%> (-9.79%)	⬇️
src/modules.h	33.33% <0%> (-66.67%)	⬇️
src/cyclotom.h	50% <0%> (-50%)	⬇️
src/records.h	50% <0%> (-50%)	⬇️
src/gapstate.h	42.85% <0%> (-42.86%)	⬇️
src/vec8bit.h	51.78% <0%> (-42.86%)	⬇️
src/macfloat.h	57.14% <0%> (-42.86%)	⬇️
src/code.h	57.14% <0%> (-40.16%)	⬇️
src/gvars.h	66.66% <0%> (-33.34%)	⬇️
... and 90 more