MAVExplorer: Add histrograms

[stephendade]

This PR adds histrogram-style graphs to MAVExplorer. This is useful for analysing the performance (such as crosstrack) of a vehicle.

Can be used similarly to graphs:
histrogram NTUN.XTrack

The histrogram will autoupdate it's range based on panning or zooming on other graphs.

There are 2 options:
-hist_bins for specifying the number of histrograms bins (default 50)
-hist_show_stats for overlaying the median and std dev (default true)

[tridge]

Automated review note — AI-generated (Claude), validated against the live diff. Please sanity-check before acting.
Full report: https://uav.tridgell.net/DevCallReviews/2026_09_01_AIReview/devcall_pr_reviews.html#prMAVProxy-1681

Reviewed at head ea41a2ea94. REQUEST CHANGES — the feature is purely additive and the process model is right; the verdict rests on two defects I reproduced, both of which make it misbehave silently rather than fail visibly.

1. A single NaN turns every overlaid statistic into NaN, and the plot still looks fine. np.median/np.mean/np.std are called without filtering, so one NaN gives median nan mean nan std nan, axvline(x=nan) draws nothing, and the legend reads "Median nan". Meanwhile matplotlib silently drops NaN inside hist(), so the bars render normally — the user gets a plausible histogram with the statistics quietly gone. Reproduced: 1000 samples + one NaN → all three stats NaN; with y = y[np.isfinite(y)] → median 0.005, mean 0.021, std 1.004. NaN is common in MAVExplorer expressions (EKF innovations, sqrt/asin out of range). One filter line at the top of _draw_histogram fixes it — and it also fixes the weighting, which currently counts NaNs in the denominator that hist then discards (bars summed to 99.9%).

2. Bad bins or non-finite data kill the child process with no message anywhere. set hist_bins 0 and histogram --bins 0 are both accepted (the MPSetting has no range=), and since plotting happens in a spawned child the exception is invisible in MAVExplorer's console — no window, no error. All three reproduced against the exact ax.hist(..., weights=...) call: bins=0 → ValueError: bins must be positive, when an integer; an inf in the data → supplied range of [...] is not finite; an all-NaN field → autodetected range of [nan, nan] is not finite.

Also worth fixing: "field has no samples" is only print()ed inside the child, so a GUI session shows nothing — and the parent already knows the series is empty before it spawns. A time-range selection with zero samples assigns self.xlim before the empty-guard returns, leaving the previous full-range histogram on screen looking current while everything else is zoomed elsewhere. Quoted expressions containing spaces are split into separate fields with only index zero plotted. The Tools > Histogram menu item dispatches with no arguments and is a no-op. And the completion rule and help string both advertise multiple fields while the command explicitly rejects them.

Checked and correct, so nobody re-litigates it: the process model is right — everything wx/pyplot-touching runs only in the child, pipe-end ownership matches Graph_UI/MavGraph exactly, and check_xlim_change returning None is correct here because the histogram's x-axis is a value axis that must not be broadcast as a time range. No figure leak. And the "histrogram" misspelling in the title did not reach any code, command name or user-visible string.