don't refer to internal variable names in gemv exceptions

[stevengj]

Currently, the DimensionMismatch exceptions refer to internal parameter names of gemv!, which is confusing because those don't necessarily match the caller's variables. For example:

julia> B = rand(1000,100); c = rand(1000);

julia> B * c
ERROR: DimensionMismatch: second dimension of A, 100, does not match length of x, 1000
Stacktrace:
 [1] gemv!
   @ ~/.julia/juliaup/julia-1.11.2+0.x64.apple.darwin14/share/julia/stdlib/v1.11/LinearAlgebra/src/matmul.jl:438 [inlined]

This PR changes the error messages to refer to "matrix" instead of "A" and "input vector" instead of "x", etc.

[mcabbott]

Seems like a good idea. It's not just gemv! though:

julia> [1 2 3; 4 5 6] * (1:10)
ERROR: DimensionMismatch: matrix A has dimensions (2,3), vector B has length 10
Stacktrace:
 [1] _generic_matvecmul!
   @ ~/.julia/dev/julia/usr/share/julia/stdlib/v1.12/LinearAlgebra/src/matmul.jl:946 [inlined]
  
julia> rand(3,4) * rand(3,4)
ERROR: DimensionMismatch: A has dimensions (3,4) but B has dimensions (3,4)
Stacktrace:
 [1] gemm_wrapper!
   @ ~/.julia/dev/julia/usr/share/julia/stdlib/v1.12/LinearAlgebra/src/matmul.jl:752 [inlined]

[stevengj]

Yeah, I knew there were more cases, but didn't have a chance to track them all down in this PR. File an issue?