fix for the multiple-decl problem - please test

[Paul Richard Thomas]

Daniel,

Apart from being continuously distracted by (a) day-time work; (b)
regressions; and

> Paul, any progress? :)

(c).... yes, I have:-)  It took more time than it should but I got rid
of the bit rot from the original patch and have it working as well or
as badly as then.

The next step is to sort out the regressions and then to implement the
improvements.  I guess that the gsymbols could be removed or used
instead of the global namespace - I have not decided which yet.

The most important thing is that there be somewhere to hang the
frontend code, the backend_decl for the procedure and the formal
argument list.

As soon as I have done PR38907, I'll quit regressions for a bit and
get on with 4.5.0 work.

Cheers

Paul