Adding jprod_residual! and jtprod_residual!

[AntoninKns]

Adding a workaround of jprod_residual! and jtprod_residual! in order for BundleAdjustmentModels to work better with other packages from JSO. The true detailed version of jprod_residual! and jtprod_residual! should be added later in order to avoid allocations and have better performance for these functions.

[codecov]

Codecov Report

Merging #64 (13fd79f) into main (aec61a2) will increase coverage by 0.44%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main      #64      +/-   ##
==========================================
+ Coverage   92.20%   92.65%   +0.44%     
==========================================
  Files           5        5              
  Lines         295      313      +18     
==========================================
+ Hits          272      290      +18     
  Misses         23       23              

Impacted Files	Coverage Δ
src/BundleAdjustmentNLSFunctions.jl	96.80% <100.00%> (+0.53%)	⬆️

---

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 aec61a2...13fd79f. Read the comment docs.

[dpo]

I don't think this is a good idea. This is going to be so much slower than what you have now.

[AntoninKns]

@tmigot asked me to add them in case these functions are called by other JSO packages at least it won't crash. I'll let him correct me, I might not have understood the reasons well.

[AntoninKns]

If you think it's worth the time I can directly implement the jprod_residual! and jtprod_residual! manually based on the jacobian description.

[tmigot]

I asked Antonin this, so that we can use the package. I get that this is a very basic version (and not efficient), but we can improve it later when Antonin is further advanced with his project.

[dpo]

Let's at least have a caching system for the Jacobian: only evaluate it if x changes. You should be able to compute the shasum of x and compare that instead of checking each component, and only be wrong with probability ε.