Hello everyone,

this patch implements module-wide gcc directive support to apply
attributes to the whole module.  Patch closely follows how plain SAVE
statement is implemented with seen_save/gfc_save_all() by using
"mod_ext_attr" bitfield in a current namespace.

There are certain complications how WEAK attribute could be safely
applied to non local symbols, applying WEAK attribute requires to
delay this until build_function_decl() sets the final attr.access
field.

For a time being I opted out to explicitly reject setting the
NO_ARG_CHECK attribute for the whole module.  There should not be a
legitimate case to apply it for every dummy argument.

One DEPRECATED diagnostic (for deprecated parameter use case inside
the same module) did not carry out from the original
gfortran.dg/attr_deprecated.f90 testcase.  Not sure how to tackle it
yet, parsing order is a bit confusing.  Giving diagnostics for
deprecated symbols inside the whole module marked as deprecated seems
to be a bit excessive.  Only giving diagnostics for the actual use
cases of such marked module components would seem to be more
intuitive.  Also from the original testcase:
use m  ! { dg-warning "Using parameter 'parm' declared at .1. is deprecated" }
shouldn't diagnostic prefer to report "use" location instead of
declaration, since it is more important for the user?  The declaration
location could be additionally reported as dg-note using some helper
function that checks if the symbol is "imported at (1)" versus
"declared at (1)".

Regtested cleanly on x86_64-pc-linux-gnu.

Regards,
Rimvydas