feat!: this is now an automatic differentiation library

[c0dearm]

Building neural networks at compile time was a very fun idea, but in practice it lacked a lot of flexibility.

For instance, you could only build sequential and static models, vs the directional acyclic and dynamic graphs that are required in most of the machine learning applications nowadays. Coding the derive macro to build the backward/reverse pass was also a feat not for the faint of heart, and I have a history!

This new approach takes the library closer to what a sane person would do, i.e. something like Tensorflow or PyTorch, and it works wonderfully!

[codecov-commenter]

Codecov Report

Merging #4 (fceea1e) into main (01ca605) will decrease coverage by 19.46%.
The diff coverage is 74.12%.

@@             Coverage Diff             @@
##             main       #4       +/-   ##
===========================================
- Coverage   93.58%   74.12%   -19.47%     
===========================================
  Files           4        7        +3     
  Lines          78      286      +208     
===========================================
+ Hits           73      212      +139     
- Misses          5       74       +69     

Impacted Files	Coverage Δ
src/context/storage.rs	10.00% <10.00%> (ø)
src/context/function.rs	46.29% <46.29%> (ø)
src/context/mod.rs	51.61% <51.61%> (ø)
src/tensor.rs	73.33% <73.33%> (ø)
src/gradient.rs	77.77% <77.77%> (ø)
src/context/tape.rs	81.81% <81.81%> (ø)
src/ops.rs	100.00% <100.00%> (ø)

---

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 01ca605...fceea1e. Read the comment docs.