Add support for creating correlated variables from a covariance or correlation matrix

[LukasACH]

This PR adds four new functions:

Measurements.covariance_matrix(n::AbstractVector{<:Measurement})

to calculate the covariance matrix from a vector of measurements,

Measurements.correlation_matrix(n::AbstractVector{<:Measurement})

to calculate the correlation matrix, and

Measurements.correlated_values(
    nominal_values::AbstractVector{<:Real},
    covariance_matrix::AbstractMatrix{<:Real}
)

and

Measurements.correlated_values(
    nominal_values::AbstractVector{<:Real},
    standard_deviations::AbstractVector{<:Real},
    correlation_matrix::AbstractMatrix{<:Real}
)

to create correlated variables from a covariance and correlation matrix respectively.

The algorithms are taken from the python package uncertatinties. If that is not okay, please tell me.

If there are issues with code styling, or if something is missing, I am happy to fix or add it.

[LukasACH]

Do you want to keep the backwards compatibility to version 1.0? If yes, I will have to find some compromises regarding readability/speed with the code.

[giordano]

I haven't looked into the details of the implementation, but this is awesome!

I'm happy to drop support for older versions of Julia, what would be the first one needed as things are now?

[giordano]

I'd appreciate documenting this in the Usage and Examples sections of the manual

[LukasACH]

I will add the corresponding usage documentation and a few examples in the coming days as soon as I find some more time.

[codecov-commenter]

Codecov Report

Merging #137 (d87a7b4) into master (d1ac981) will increase coverage by 0.18%.
The diff coverage is 100.00%.

❗ Current head d87a7b4 differs from pull request most recent head 45db0d4. Consider uploading reports for the commit 45db0d4 to get more accurate results

@@            Coverage Diff             @@
##           master     #137      +/-   ##
==========================================
+ Coverage   95.45%   95.63%   +0.18%     
==========================================
  Files          12       12              
  Lines         726      756      +30     
==========================================
+ Hits          693      723      +30     
  Misses         33       33              

Impacted Files	Coverage Δ
src/Measurements.jl	86.66% <ø> (ø)
src/utils.jl	100.00% <100.00%> (ø)

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[LukasACH]

I have added documentation for both, could you check if it fits your expectation?

Also, are you happy with the function names? There are existing functions in Statistics named cov and cor for calculating the covariance and correlation matrices. Would it be better to extend these?

[giordano]

BTW, this PR solves the only point in the TODO list

Measurements.jl/docs/src/todo.md

Lines 8 to 12 in d417519

	TODO
	----
	* Allow defining quantities related by a correlation matrix and correctly
	propagate uncertainty in this case

which can now be removed 🙂

[LukasACH]

Thanks for your comments, I will fix them.

What is your opinion on extending the functions in Statistics named cov and cor for calculating the covariance and correlation matrices instead of creating our own?
There are other packages like Distributions.jl which extend these functions.
If we extend Statistics, should we export cov and cor or let the user import Statistics themselves?

[giordano]

Thanks a lot for this nice contribution, much appreciated and awaited!