It seems that there is a code generation bug in GCC which has crept into
our binaries. For zero-length ranges near zero, the integer manipulation
logic does not work and comes out with a zero range again.

Debugging shows that the floating point values come out of the
AdjustTRange function fine, but are zero'd again when assigned back into
the double using the cvtss2sd instruction.

This reverts the following commits:

    "Add include needed for old gcc versions"
    commit d93bc11b

    "Update the tiny range delta to work when passing in doubles"
    commit fe2cf0b0

    "Fix strict aliasing warnings in vtkSMCoreUtilities AdjustRange"
    commit 450a1545

    "vtkSMCoreUtilities now adjusts ranges properly when min~=max."
    commit 604cbe44