Inline blocksizes for better type-inference

[jishnub]

Fixes #425

The type-inference should work within a function now, as the dimension will be propagated to axes as a constant:

julia> using BlockArrays, Test

julia> x = BlockedArray(rand(4, 4), [2, 2], [2, 2]);

julia> @inferred (x -> blocksizes(x, 1))(x)
2-element Vector{Int64}:
 2
 2

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 93.55%. Comparing base (a62bd8e) to head (1d2c7d0).
Report is 3 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #429   +/-   ##
=======================================
  Coverage   93.55%   93.55%           
=======================================
  Files          18       18           
  Lines        1644     1644           
=======================================
  Hits         1538     1538           
  Misses        106      106           

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

[lkdvos]

I can confirm that this resolved the type-instability in #426 , thank you!

As a small comment, that I rejected myself after going through the actual implications, but might want to leave here for future reference.

As an additional suggestion, this might also not be unreasonable, and further improve the type-stability in the case of block arrays with BlockedOneTo axes:

Currently, the combination of this definition and this definition imply that the following things don't coincide:

a = rand(3)
blockaxes(a, 1) # BlockRange(Base.OneTo(1))
blockaxes(a, 2) # Base.OneTo(1)
blockaxes(axes(a, 2), 1) # BlockRange(Base.OneTo(1))

To me it seems like the latter is a much more reasonable fallback for "trailing indexes", and this would resolve the type-instability for the typical BlockArray cases.

This suggestion does not really work because I didn't fully realize that a[Block(1), 1] and a[Block(1), Block(1)] do not actually mean the same thing (the former is a vector, the latter a matrix).