Fix semantic analysis failures of LBOUND/UBOUND in type specification

[wanbinchen-hnc]

An issue occurs when LBOUND/UBOUND appears in the type specification
of a function result and the DIM argument is missing.

The root cause of this issue is that LBOUND/UBOUND is converted to
runtime call in semantic phase, which is too early for type-spec of
a function result. On the calling side of the function, we have no
way of knowing whether LBOUND/UBOUND exists or its reference to the
function's dummy arguments.

We fix the issue by delaying the conversion to the tranform phase.
In particular, we perform the conversion in dummy-actual argument
replacement if the array argument of LBOUND/UBOUND is an assumed-shape
array dummy.

In addition, we fix an ICE that occurs when LBOUND/UBOUND appears in
subscript expressions in type-spec, fix the link error that occurs
when both DIM and KIND arguments are present, fix an issue with LBOUND
on the last dimension of an assumed-size array, and add support for
LBOUND/UBOUND on an assumed-rank array.

[bryanpkc]

There are typos and format errors in your commit message; please reword it. It should start with a commit title, with a topic prefix, and separated from the message body with a blank line. For example:

[flang1] Fix handling of LBOUND/UBOUND in type specification

An issue occurs when LBOUND/UBOUND appears in the type specification
of a function result and the DIM argument is missing.

The root cause of this issue is that LBOUND/UBOUND is converted to
runtime call in semantic phase, which is too early for type-spec of
a function result. On the calling side of the function, we have no
way of knowing whether LBOUND/UBOUND exists or its reference to the
function's dummy arguments.

We fix the issue by delaying the conversion to the tranform phase.
In particular, we perform the conversion in dummy-actual argument
replacement if the array argument of LBOUND/UBOUND is an assumed-shape
array dummy.

In addition, we fix an ICE that occurs when LBOUND/UBOUND appears in
subscript expressions in type-spec, fix the link error that occurs
when both DIM and KIND arguments are present, fix an issue with LBOUND
on the last dimension of an assumed-size array, and add support for
LBOUND/UBOUND on an assumed-rank array.

[wanbinchen-hnc]

@bryanpkc Thanks for your comments, I have updated the commit message.

[pawosm-arm]

I can give you a tentative approval. This is an extensive change where it would be difficult to prove validity of the consequences of all the possible interactions. Therefore I've hauled this commit through various flows of our CI, and it didn't break any of the workloads, also NAG tests seemed to be happy with it.

[bryanpkc]

This has passed our internal testing.