Use H5Pisa_class to circumvent bug
Fixes #218. On the the bug in the issue memory usage decreases from 100MB++ to less than 50kB
@stormmathisen and @smutch Would be great if you tested this
Hi @mulimoen. Thanks for looking in to this.
I can confirm a noticeable improvement for my small test run (3,819 groups with ~11,500 datasets).
Before:

After:

There is a large reduction in persistent 96 byte allocations.
- Probably makes sense to fix a few clippy complaints as well here?
- Add this to changelog?
- Why do you say it "partially" fixes #218? Wasn't this the root of the problem?
There are enough breaking clippy complaints that I've opened #220 for those.
Added an entry to the changelog and changed the description of this PR
Rebase on master?
👍