Cuda MAP/Sample on new bit circuits

[khosravipasha]

1. Adds to bitrcuit the begining and end of edge ids for each node.
2. Make upward customizable with functions for input_init_func and sum_agg_func. Since we initialize input nodes and handle sum nodes differently for MAP
3. Downward pass for MAP (preprocess map state and map-lls for each input on the bpc.heap, then use parallel stacks to keep track of choices until input nodes)

Todos:

• Adding tests
• Tests comparing cpu and gpu versions
• Test on hybrid input types to see if kernels still compile
• Somehow add tests for sample (maybe with a fixed random seed)

[codecov]

Codecov Report

Merging #116 (7d3fb13) into master (c98b568) will decrease coverage by 5.38%.
The diff coverage is 4.41%.

@@            Coverage Diff             @@
##           master     #116      +/-   ##
==========================================
- Coverage   47.84%   42.46%   -5.39%     
==========================================
  Files          18       20       +2     
  Lines        1417     1731     +314     
==========================================
+ Hits          678      735      +57     
- Misses        739      996     +257     

Impacted Files	Coverage Δ
src/abstract_nodes.jl	87.75% <0.00%> (-1.83%)	⬇️
src/queries/likelihood.jl	2.18% <0.00%> (-0.04%)	⬇️
src/queries/map.jl	0.00% <0.00%> (ø)
src/queries/sample.jl	0.00% <0.00%> (ø)
src/structures/rat.jl	0.00% <0.00%> (ø)
src/input_distributions.jl	29.46% <4.25%> (-18.30%)	⬇️
src/bits_circuit.jl	87.70% <77.77%> (-0.64%)	⬇️
src/queries/sample_cpu.jl	100.00% <100.00%> (+100.00%)	⬆️
src/queries/likelihood_cpu.jl	58.33% <0.00%> (-1.29%)	⬇️

---

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 c98b568...7d3fb13. Read the comment docs.