[FEA] Support resize/reshape for mdarray.

[trivialfis]

As suggested by @rg20 and @cjnolet in #437, some helper functions for late initialization of mdarray and support for reshape will greatly improve the usability of this new abstraction.

[cjnolet]

Just leaving a note that this is going to be needed before building our new sparse sparse formats.

[github-actions]

This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.

[github-actions]

This issue has been labeled inactive-90d due to no recent activity in the past 90 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed.

[mhoemmen]

Regarding mdarray (the C++ Standard Library proposal P1684), I posted some discussion here on why mdarray does not include this feature: ORNL/cpp-proposals-pub#287

[trivialfis]

Thank you for the discussion. I think reshaping without the size being changed would be really useful. As for changing the size, I haven't seen any actual use case other than changing the size from 0 to something else (initializing a variable). @cjnolet @divyegala might be able to provide more context.

[mhoemmen]

@trivialfis Thanks for taking a look at this!

I think reshaping without the size being changed would be really useful.

Does this mean taking an mdspan, and a new layout mapping or extents, and returning an mdspan viewing a subset of the input mdspan's elements with the new layout mapping?

[trivialfis]

That sounds right except for the subset part, which should be handled by submdspan instead of reshape. I think changing the mapping should be sufficient. From my perspective, it would be as simple as taking a vector of size 4 and returning a matrix of shape 2x2. No need for changing size.

[github-actions]

This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.