EventProp fixes

[neworderofjamie]

This branch contains several EventProp fixes that have emerged from my recent debugging deep dive:

• LIF + ExpCurr neuron model were not matching Thomas's code with exact integration. Slightly extended ValueDescriptor system to enable multiple derived GeNN variables to be calculated from a single Python parameter to tidily implement scale factor already present in GeNN ExpCurr model. This is configured via the standard compiler default_params mechanism.
• I think because we reset time every batch, the ring buffer is not empty at the start of each epoch like it is in Thomas's code. To address this we move the new gradient zeroing update to the end of batch 1 to ensure gradients are accumulated correctly.
• Rather than using the per-neuron spike counts for regularisation, Thomas's code reduced these over all batches. This functionality has now been implemented,

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 40.00000% with 39 lines in your changes missing coverage. Please review.

Project coverage is 64.77%. Comparing base (b729bbd) to head (c522e76).

Files with missing lines	Patch %	Lines
ml_genn/ml_genn/compilers/event_prop_compiler.py	19.35%	25 Missing :warning:
ml_genn/ml_genn/synapses/exponential.py	36.36%	7 Missing :warning:
ml_genn/ml_genn/utils/value.py	63.15%	7 Missing :warning:

Additional details and impacted files

```diff @@ Coverage Diff @@ ## master #115 +/- ## ========================================== - Coverage 65.17% 64.77% -0.41% ========================================== Files 101 101 Lines 4221 4261 +40 ========================================== + Hits 2751 2760 +9 - Misses 1470 1501 +31 ```

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.