Use function barriers for MulAddAccumulate

[DanielVandH]

Relies on JuliaArrays/LazyArrays.jl#386. Don't think the performance in the simple benchmark below is affected by JuliaArrays/LazyArrays.jl#383

Puts the MulAddAccumulate's cache_filldata! method behind a function barrier, cutting the number of allocations roughly in half for a very simple example. On current release:

using LazyArrays, SemiclassicalOrthogonalPolynomials, BenchmarkTools, InfiniteArrays
using SemiclassicalOrthogonalPolynomials: MulAddAccumulate
@benchmark copyto!(dest, src) setup=begin
    A = Vcat(1, cache(2:3), cache(4:∞))
    B = Vcat(cache(1:5), [1, 2], cache(3:∞))
    src = view(MulAddAccumulate(A, B), 1:100_000)
    dest = similar(src)
end

BenchmarkTools.Trial: 123 samples with 1 evaluation per sample.
 Range (min … max):  36.005 ms … 72.708 ms  ┊ GC (min … max): 0.00% … 3.77%
 Time  (median):     38.956 ms              ┊ GC (median):    4.51%
 Time  (mean ± σ):   40.786 ms ±  5.312 ms  ┊ GC (mean ± σ):  4.01% ± 2.82%

      ▂█▁
  ▃▄▃▃███▅▄▃▃▆▂▃▂▁▂▃▂▁▂▂▁▂▃▂▁▁▂▁▁▁▁▂▁▁▁▁▁▁▁▁▁▂▁▁▁▁▁▁▁▁▁▁▂▁▁▁▂ ▂
  36 ms           Histogram: frequency by time          65 ms <

 Memory estimate: 19.16 MiB, allocs estimate: 797057.

On this PR:

BenchmarkTools.Trial: 142 samples with 1 evaluation per sample.
 Range (min … max):  31.669 ms … 48.247 ms  ┊ GC (min … max): 0.00% … 5.63%
 Time  (median):     35.093 ms              ┊ GC (median):    6.51%
 Time  (mean ± σ):   35.255 ms ±  2.779 ms  ┊ GC (mean ± σ):  4.34% ± 3.45%

     ▄▁▂      ▇██▅
  ▄▃▆████▅▃▅▅▆█████▅▄▃▃▃▃▁▅▁▄▁▁▃▄▃▃▁▁▁▁▃▁▁▁▁▁▁▁▁▁▃▁▁▁▃▁▁▁▁▁▁▄ ▃
  31.7 ms         Histogram: frequency by time          47 ms <

 Memory estimate: 17.68 MiB, allocs estimate: 399625.

Doesn't really end up making as much of a difference inside divdiff for a HalfWeighted{:a} polynomial though, e.g.

@benchmark D[1:n, 1:n] setup = begin
    P = SemiclassicalJacobi(2.0, 1 / 2, 0, 1 / 2);
    Q = SemiclassicalOrthogonalPolynomials.HalfWeighted{:a}(P);
    D = diff(Q).args[2];
    n = 1000;
end

gives roughly the same results. Still need to make some performance improvements there as discussed.