Respect type of parameters in fit for Bernoulli, Binomial, and Uniform

[devmotion]

As a follow-up to #1551 and #1557, this PR addresses #1544 for Bernoulli, Binomial, and Uniform.

With this PR fit(Uniform{Float32}, ...) etc. will actually estimate a distribution of type Uniform{Float32}, regardless of the observations. One can still use fit(Uniform, ...) if the parameter type should not be fixed but depend on the type of the observations.

[codecov-commenter]

Codecov Report

Merging #1558 (7699b7b) into master (a350622) will not change coverage.
The diff coverage is 92.85%.

@@           Coverage Diff           @@
##           master    #1558   +/-   ##
=======================================
  Coverage   85.52%   85.52%           
=======================================
  Files         128      128           
  Lines        7863     7863           
=======================================
  Hits         6725     6725           
  Misses       1138     1138           

Impacted Files	Coverage Δ
src/univariate/discrete/binomial.jl	94.28% <88.88%> (ø)
src/univariate/continuous/uniform.jl	92.20% <100.00%> (ø)
src/univariate/discrete/bernoulli.jl	89.39% <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 a350622...7699b7b. Read the comment docs.

[matbesancon]

I agree with this in principal, it was either this solution or a promotion mechanism, but using the specified type param makes more sense when provided

[gdalle]

Thank you for this @devmotion !