Loss recording

[neworderofjamie]

After some thought and fiddling, my conclusion is that loss is a compiler-specific thing and needs calculating in the backward pass so recording it isn't really something that can be easily or sensibly implemented as a metric. Therefore, this PR adds functionality to generate loss-calculating terms to the event prop backward pass and stores the data in a callback added by the compiler (in the same way rewiring stats are stored with Deep-R in #94).

However, I cannot quite figure out how all the calculations works in the original genn_eventprop code

[codecov-commenter]

Codecov Report

❌ Patch coverage is 59.61538% with 21 lines in your changes missing coverage. Please review.
✅ Project coverage is 77.40%. Comparing base (24c3282) to head (1bef0b7).

Files with missing lines	Patch %	Lines
ml_genn/ml_genn/compilers/event_prop_compiler.py	53.48%	20 Missing ⚠️
..._genn/ml_genn/losses/relative_mean_square_error.py	66.66%	1 Missing ⚠️

Additional details and impacted files

@@              Coverage Diff              @@
##           ml_genn_3     #183      +/-   ##
=============================================
- Coverage      77.56%   77.40%   -0.17%     
=============================================
  Files            114      114              
  Lines           4788     4833      +45     
=============================================
+ Hits            3714     3741      +27     
- Misses          1074     1092      +18     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.