[npazosmendez]

Fixes #523 and the one-step time shift of the movingMedian output.

movingMedian had its own implementation, with these bugs:

• The output started one step too early (at from, not at from + step). When you combine it with an aligned series, for example asPercent(x, movingMedian(x, '1h')), the first point and the last point are null.
• Integer windows did not fetch earlier data, so the first windowSize - 1 points were null.
• Windows with NaN values gave wrong medians, because github.com/JaderDias/movingmedian does not handle NaN.
• xFilesFactor was ignored.

The moving package already had a median branch without these bugs (most fixes came from #783). But movingMedian was not registered there, so the branch was never used. This PR registers movingMedian in moving, and removes the movingMedian package and the github.com/JaderDias/movingmedian dependency.

Changes to moving on top of that:

• Speed: Windowed.Median() copied and sorted the window for each point. It now uses two indexed heaps that skip NaN values, so each push is O(log window).
• Speed: the xFilesFactor check uses the NaN count of the window, so it does not scan the full window for each point.
• Fix: when returnNaNsIfStepMismatch is false, data that is returned unchanged now keeps its timestamps. Before, they moved forward by the window size.

The expected values in the new tests come from graphite-web (movingWindow with median).

movingMedian now reads returnNaNsIfStepMismatch from the moving function config. A functionsConfig.movingMedian entry has no effect now.

A couple of unrelated fixes follow in #950

Benchmark

Benchmarks show how the new is a bit slower, but not by orders of magnitude (on the flip side, it's more memory efficient). The trade is correctness. I'm sure this can be optimized further, but it's not my intent here.

goos: linux
goarch: amd64
pkg: github.com/go-graphite/carbonapi/expr/types
cpu: Intel(R) Core(TM) Ultra 7 165U
                        │     old      │                  new                  │
                        │    sec/op    │    sec/op      vs base                │
WindowedMedian/4-14       420.5µ ± 14%    810.2µ ± 30%  +92.69% (p=0.000 n=10)
WindowedMedian/60-14      848.6µ ± 12%   1682.0µ ± 20%  +98.21% (p=0.000 n=10)
WindowedMedian/360-14     1.277m ± 26%    2.489m ±  9%  +94.87% (p=0.000 n=10)
WindowedMedian/600-14     1.539m ± 10%    2.799m ± 17%  +81.89% (p=0.000 n=10)
WindowedMedian/3600-14    2.430m ± 10%    3.742m ± 10%  +53.99% (p=0.000 n=10)
WindowedMedian/60480-14   18.04m ±  7%    19.92m ± 11%        ~ (p=0.063 n=10)
geomean                   1.770m          2.985m        +68.65%

                        │     old      │                 new                  │
                        │     B/op     │     B/op      vs base                │
WindowedMedian/4-14         336.0 ± 0%     288.0 ± 0%  -14.29% (p=0.000 n=10)
WindowedMedian/60-14      2.172Ki ± 0%   3.000Ki ± 0%  +38.13% (p=0.000 n=10)
WindowedMedian/360-14     14.92Ki ± 0%   18.34Ki ± 0%  +22.93% (p=0.000 n=10)
WindowedMedian/600-14     27.92Ki ± 0%   30.34Ki ± 0%   +8.67% (p=0.000 n=10)
WindowedMedian/3600-14    158.4Ki ± 0%   184.1Ki ± 0%  +16.20% (p=0.000 n=10)
WindowedMedian/60480-14   3.413Mi ± 0%   2.797Mi ± 0%  -18.04% (p=0.000 n=10)
geomean                   23.40Ki        25.06Ki        +7.07%

                        │     old     │                new                 │
                        │  allocs/op  │ allocs/op   vs base                │
WindowedMedian/4-14        8.000 ± 0%   5.000 ± 0%  -37.50% (p=0.000 n=10)
WindowedMedian/60-14      15.000 ± 0%   5.000 ± 0%  -66.67% (p=0.000 n=10)
WindowedMedian/360-14     21.000 ± 0%   5.000 ± 0%  -76.19% (p=0.000 n=10)
WindowedMedian/600-14     23.000 ± 0%   5.000 ± 0%  -78.26% (p=0.000 n=10)
WindowedMedian/3600-14    29.000 ± 0%   5.000 ± 0%  -82.76% (p=0.000 n=10)
WindowedMedian/60480-14   49.000 ± 0%   5.000 ± 0%  -89.80% (p=0.000 n=10)
geomean                    20.86        5.000       -76.03%