TEST: Numpy changed longdouble str representations in 1.18

[effigies]

numpy/numpy#10151 broke the h5py compatibility tests which checked for string matching.

That wasn't an important test, but it did explain why the issue took so long to track down. This update adds numpy 1.18's changes to the logic.

All of this will go away when we drop Python 3.5.

[codecov]

Codecov Report

Merging #858 into master will decrease coverage by 0.2%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master     #858      +/-   ##
==========================================
- Coverage   90.07%   89.87%   -0.21%     
==========================================
  Files          96       96              
  Lines       12392    12392              
  Branches     2186     2186              
==========================================
- Hits        11162    11137      -25     
- Misses        881      906      +25     
  Partials      349      349

Impacted Files	Coverage Δ
nibabel/keywordonly.py	92.85% <0%> (-7.15%)	⬇️
nibabel/nicom/dicomwrappers.py	85.16% <0%> (-5.75%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update a66c0b2...fadfb45. Read the comment docs.

[skoudoro]

By curiosity, when do you plan do drop python3.5?

[effigies]

By curiosity, when do you plan do drop python3.5?

First minor release after March 14, 2020: #803

[skoudoro]

Thank you for this info and the pointer to #803

[effigies]

Failed test to be resolved in #857.

[effigies]

Merged into master, not maint/3.0.x. Cherry-picked into maint/3.0.x in 50abcb3.