Rework rejection sampling

[mj-will]

This PR reworks the rejection sampling in the proposal classes so they accumulate samples until sufficient samples are obtained, rather than apply rejection sampling to each batch of samples. This should improve the estimate of the normalisation constant.

Main changes

• Rework FlowProposal.populate

Other changes
Whilst making these changes, I made a couple of other minor changes:

• The population time no longer includes the time spent evaluating the log-likelihood
• FlowProposal.compute_weights no longer normalises the weights and can also return the value of the log-prior rather than updating the samples in place
• The population acceptance is now computed based on all the accepted samples rather than the N final samples. Previously, this meant that the reported acceptance could be lower than the actual value.
• Added a deprecation warning to FlowProposal.rejection_sampling since it is no longer used.
• FlowProposal.plot_pool no longer takes z as argument
• Refactor some tests

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (b389742) 93.29% compared to head (992dd43) 93.32%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #358      +/-   ##
==========================================
+ Coverage   93.29%   93.32%   +0.02%     
==========================================
  Files          67       67              
  Lines        6327     6338      +11     
==========================================
+ Hits         5903     5915      +12     
+ Misses        424      423       -1     

Flag	Coverage Δ
unittests	93.32% <100.00%> (+0.02%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

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