Core 2D binning routine for multiple functions

[leewujung]

A few functions require the same binning routine along ping_time and depth or along distance and depth: compute_MVBS, compute_NASC, estimate_noise. This is a simple operation if this binning is based on indices (e.g., 0, 1, 2, ...), but becomes non-trivial if the binning is based on the labels (e.g., the actual ping_time which are not integers). This is because in the latter case checking the actual values in the labels is required to determine which bin a data point should fall in.

The label spacing along the depth dimension is typically the same when the data is calibrated using a single sound speed value, since the sample_interval is constant within a single ping. However, a case that sometimes occurs is when the sample_interval switches to another value after a certain number of pings. This is especially likely to occur if we combine EchoData from multiple files. Doing binned averages for such as case was added in #878, but there continues to be efficiency/memory issue related to the implementation.

This issue is for working on a core 2D binning routine to enhance efficiency, memory usage, and robustness against different types of data.

Related issues:

• Further improve compute_MVBS in computation across ping time and channel #888
• Memory consumption and accumulation, and performance issues, in v0.6.3 #872
• Rework MVBS tests #756
• Further improve 'compute_NASC' to use flox #1127