WIP: Efficient inference for regularly-spaced observations under the posterior

[willtebbutt]

Currently inference in the posterior is a bit sluggish because it's not possible to properly exploit regular spacing under the posterior. This PR implements the data structures and code necessary to attain efficient posterior inference.

[codecov[bot]]

Codecov Report

Merging #33 (a434d86) into master (4bdda0d) will decrease coverage by 65.59%. The diff coverage is 12.79%.

@@             Coverage Diff             @@
##           master      #33       +/-   ##
===========================================
- Coverage   89.95%   24.35%   -65.60%     
===========================================
  Files          20       19        -1     
  Lines         896      817       -79     
===========================================
- Hits          806      199      -607     
- Misses         90      618      +528     

Impacted Files	Coverage Δ
src/TemporalGPs.jl	100.00% <ø> (ø)
src/models/checkpointed_immutable_pullbacks.jl	0.00% <0.00%> (-87.29%)	:arrow_down:
src/models/immutable_inference.jl	52.83% <ø> (-47.17%)	:arrow_down:
src/models/immutable_inference_pullbacks.jl	0.00% <0.00%> (-96.59%)	:arrow_down:
src/util/zygote_rules.jl	0.81% <0.00%> (-61.09%)	:arrow_down:
src/gp/posterior_lti_sde.jl	79.06% <60.00%> (-7.42%)	:arrow_down:
src/util/regular_data.jl	58.82% <62.50%> (-30.07%)	:arrow_down:
src/util/mul.jl	0.00% <0.00%> (-100.00%)	:arrow_down:
src/space_time/to_gauss_markov.jl	0.00% <0.00%> (-100.00%)	:arrow_down:
src/space_time/rectilinear_grid.jl	0.00% <0.00%> (-100.00%)	:arrow_down:
... and 14 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update c5e7cfa...a434d86. Read the comment docs.