[F32] Fixes to translate test infrastructure & metric

[FlorianDeconinck]

PR #67 up-skilled the translate test to be able to do 32-bit float and introduces a new metric named MultiModalMetric to be able to measure differences at 32-bit float via a mix of different metrics.

This PR aim to fix a couple of missed bugs

• Fix precision when using make_storage_data. We rely know on the input array being correct instead of trying to fix them. Any incorrect data type should be dealt with in the serialbox side OR as a verbose cast in the test itself.
• Fix the logic surrounding NaN for the multimodal metric. New comment spells it out now

# Combine all distances into sucess or failure
# Success =
# - no unexpected NANs (e.g. NaN in the ref MUST BE in computation) OR
# - absolute distance pass OR
# - relative distance pass OR
# - ulp distance pass

Checklist:

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• Any dependent changes have been merged and published in downstream modules
• New check tests, if applicable, are included