ITS: GPU: prepare to lazy loading of data

[f3sch]

needs #14583.

Prepares to start processing from multiple threads e.g. lazy loading of data & computation but does not yet implement it.

• Adds needs synchronisation scaffolding
• moves loading of data to be able to it per layer and not all at once
• adds tracking of which memory was pinned (in case we drop the processing we should only unpin those that were registered)

For now this does not change how the processing is done.

The later goal for reference is to start launching kernels as soon as their data is available, e.g. getting these to overlap as best as possible.

[github-actions]

REQUEST FOR PRODUCTION RELEASES:
To request your PR to be included in production software, please add the corresponding labels called "async-" to your PR. Add the labels directly (if you have the permissions) or add a comment of the form (note that labels are separated by a ",")

+async-label <label1>, <label2>, !<label3> ...

This will add <label1> and <label2> and removes <label3>.

The following labels are available
async-2023-pbpb-apass4
async-2023-pp-apass4
async-2024-pp-apass1
async-2022-pp-apass7
async-2024-pp-cpass0
async-2024-PbPb-apass1
async-2024-ppRef-apass1
async-2024-PbPb-apass2
async-2023-PbPb-apass5