Fix generic mul accidental type promotion Int32 -> Int64

[haampie]

Fixes JuliaLang/LinearAlgebra.jl#704.

[dkarrasch]

LGTM. Did you check whether there are more unintended 0s (instead of zeros) in similar contexts?

[haampie]

matmul.jl is clear of 0 and 1 literals in computations as far as I can see.

[haampie]

😕 the current generic_mul! is really broken it seems :(

Type deduction for the output vector is way too generic:

julia> Real[1 2] * Real[1.5; 2.0]
1-element Array{Any,1}:
 5.5

vs

julia> Real[1 2] .* Real[1.5; 2.0]
2×2 Array{Float64,2}:
 1.5  3.0
 2.0  4.0

The test failure is because I'm using zero(R) where R === Any. But there's 3 other places where this is used at the moment that apparently aren't covered by tests.

Not sure what to replace zero(R) with at the moment.

[KristofferC]

Not sure what to replace zero(R) with at the moment.

Does false work?

[dkarrasch]

I'll rerun CI (the only failure seems unrelated) and merge if everything runs smoothly.