incorrect module identification for fd's #=/2 (0962)

There seem to be a problem identifying the module for some fd predicates,
at least during tracing:

Version 5.3 #14, Tue Aug 28 18:22 2001
[eclipse 1]: lib(fd).
...
[eclipse 2]: [user].
 f(X) :- X + 3 #= Y.

user       compiled traceable 148 bytes in 0.00 seconds

Yes (0.00s cpu)

The #=/2 needs to be sufficiently complicated. Something like X #= Y would
not cause this error.

[eclipse 4]: f(X).
  (1) 1 CALL  f(X)   %> creep
  (2) 2 CALL  3 - Y + X#=0   %> spy
accessing an undefined procedure from set_flag(fd_eq/1, spy, on) in module eclipse
  (2) 2 LEAVE  ... #= 0   %>

The system thinks that the goal is in module eclipse, so the attempt to
place a spy point fails. This also happens in TkECLiPSe (which is where I
actually discovered the problem)

The problem also occurs with the installed version of ECLiPSe 5.2 (on Linux
at least).