add ForwardDiff extension

[longemen3000]

Ports the code from https://github.com/longemen3000/ForwardDiffOverMeasurements.jl into an extension.

This PR only takes care of ambiguous operations (like +(x::Measurement,y::Dual)). It does not require any change in hierarchy (Measurements.Measurement isa AbstractFloat and ForwardDiff.Dual isa Real).

What is missing is some methods defined exclusively for Measurements ((like Measurements.value and Measurements.measurement), as well as constructor functions, but that discussion and PR can come later. I added the definitions for value/uncertainty (just map those over the partials), and measurement (the derivative respect to the value is dx * 1 ± 0, the derivative respect to the uncertainty is dx * 0 ± 1)

What this PR allows is allowing code that uses AD via ForwardDiff to use Measurements as primal type.

related: longemen3000/ForwardDiffOverMeasurements.jl#3

Adds a solution to #154

[codecov]

Codecov Report

Attention: Patch coverage is 96.87500% with 3 lines in your changes missing coverage. Please review.

Project coverage is 95.69%. Comparing base (ece3021) to head (2a41c0c).

Files with missing lines	Patch %	Lines
ext/MeasurementsForwardDiffExt.jl	96.84%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #178      +/-   ##
==========================================
+ Coverage   95.54%   95.69%   +0.14%     
==========================================
  Files          15       16       +1     
  Lines         764      860      +96     
==========================================
+ Hits          730      823      +93     
- Misses         34       37       +3     

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[longemen3000]

errors in nightly seem unrelated

[giordano]

Any chance of either increasing code coverage or removing the untested functions? I'm not particularly familiar with ForwardDiff internals, there's lots of code I have no idea what's doing and seeing it's not even tested makes me a bit uncomfortable 🙂

[longemen3000]

I can try. I'm gonna add some comments too. if you want some peace of mind, the code in this PR was already used some time ago in the making of this paper

[longemen3000]

Some updates:

While testing for all possible combinations of numbers, Duals and measurements, i found #179. This PR adds a fix to that issue because it is necessary for the ForwardDiff tests. Also added some explanations about what does each code section do.

[giordano]

To be clear, if JuliaDiff/ForwardDiff.jl#665 was merged we wouldn't need all these methods? It really sounds like that should be fixed upstream, not a fan of peering into internals of other packages to fix their bugs.

[longemen3000]

Yes

[dalum]

What is missing from this PR? Is any help needed?

[giordano]

I'd prefer JuliaDiff/ForwardDiff.jl#665 to be resolved upstream rather than worked around here, so helping that PR would help this one in turn.