depends.py should set unique configuration names for the outcome file

[gilles-peskine-arm]

We don't get good reporting through the outcome file for testing done via depends.py. The problem is that depends.py doesn't set the different configuration field, so all entries in the outcome file end up with just the all.sh component name in the configuration field. Outcome analysis assumes that there is a single outcome line for a given combination of {platform, configuration, test suite, test case name}.

This is a regression when moving from the separate perl scripts to the combined depends.py. It happened because we didn't yet have outcome files when we started work on depends.py, and we missed this addition when we finalized it.

The goal of this issue is for depends.py to give each configuration a unique name and set the environment variable MBEDTLS_TEST_CONFIGURATION, similar to what test-ref-configs.pl does and what the old scripts (curves.pl, etc.) did. The names going forward don't have to be exactly the same as the old names, they just need to be non-ambiguous and sensible.