"unable to commit memory" is rarely a compiler bug, update the error message

We're getting a lot of bug reports for this message:

• #17038 (closed) (host system has 1GB RAM)
• #16613 (closed)
• (probably) #16460 (closed)
• #15612 (closed) (host system has 1GB RAM)
• #15500 (closed) (the reported realized that they're running out of memory, reporter suggests updating the error message)
• #15174 (closed) (the program has huge residency -- fills 16G + more)

The error in these tickets is basically "out of memory", but we ask the user to report a bug. Instead we should at least mention the possibility of memory shortage.

Of course there are valid GHC bugs (e.g. because it allocates excessively, #15054 (closed)), but as far as I can see that's a rare event.

(A counter example is #16676 (closed), which was actually a RTS bug, probably a duplicate of #15581 (closed), but we thought the user is running out of memory)