DNM Filters: Demonstrate failures in can_apply and set_local

[crusaderky]

Unit test and reproducer for

• filters: can_apply does nothing for optional filters #6161
  H5Z_FLAG_OPTIONAL causes a return value of 0 from can_apply to be completely ignored and to blindly execute the set_local and filter callbacks afterwards.
• can_apply and set_local filter callbacks skipped for H5T_VARIABLE dtypes #5942
  vlen dtypes cause both can_apply and set_local to be completely skipped. This breaks blosc and blosc2.

AI disclaimer

I used AI to produce the initial draft of this PR. I then proceeded to thoroughly audit and tweak the changes by hand.

---

Important

Add tests for can_apply and set_local callbacks in dataset filters, introducing new filters and verifying behavior for variable-length datasets.

• Tests:
  • Add test_can_apply_skips_cbs and test_can_apply_skips_cbs_vlen to verify can_apply callback skips set_local and filter callbacks when returning 0.
  • Add test_set_local_updates_cd and test_set_local_updates_cd_vlen to verify set_local callback updates cd_values before filter execution.
• Filters:
  • Introduce H5Z_CAN_APPLY_BOGUS4 and H5Z_SET_LOCAL_UPDATE_CD_VALUES filters in dsets.c.
  • Implement can_apply_bogus4, set_local_bogus4, filter_bogus4, set_local_update_cd_values, and filter_update_cd_values functions.
• Constants:
  • Add H5Z_FILTER_BOGUS4 and H5Z_FILTER_UPDATE_CD_VALUES constants.
  • Define UPDATED_CD_VALUE for set_local_update_cd_values.

^{This description was created by for 7e0dede. You can customize this summary. It will automatically update as commits are pushed.}

[crusaderky]

There is a lot of redundant boilerplate from this point on.
This was done to keep tests as stupid as possible.
Please advise if you would instead prefer to deduplicate the boilerplate through helper functions.