first draft of layer-as-tile code

[trautmane]

This code is intended to "roughly-3D-align" the 2D-stitch-only solve results to reduce the amount of work needed to run the results through SOFIMA. It copies/reuses many of the components developed for our MFOV-as-tile process.

The process works with and creates a set of stacks like this:

w61_s109_r00_gc_par_crc_aso :             starting 2D-stitch-only SFOV tile stack
w61_s109_r00_gc_par_crc_aso_lat :         layer-as-tile stack with dynamically rendered layer images
w61_s109_r00_gc_par_crc_aso_latr :        layer-as-tile stack with persisted layer images
w61_s109_r00_gc_par_crc_aso_latr_align :  aligned layer-as-tile stack with persisted layer images
w61_s109_r00_gc_par_crc_aso_3d :          rough 3D aligned SFOV tile stack

The process also creates a layer-as-tile match collection that is used for the layer-as-tile solve.

Our hope is that the ..._aso_3d result stacks are good enough that we can skip cross-layer SFOV match data generation and that the SOFIMA setup work will be reduced/simplified.

A few final notes ...

• We originally thought the layer-as-tile stacks might be useful for storing layer-based histogram intensity correction information, but that process has changed. We currently do not need to store histogram data.
• The layer-as-tile stacks might be useful if we want to orient the slabs similarly before we try to connect them into a single volume.

[minnerbe]

Thanks, @trautmane! Except for some small comments (and an optional offline conversation), this looks good to me.

[minnerbe]

These variables could be doubles. The only usage is a few lines below, where they're explicitly cast to double again.

[minnerbe]

I have very minor and abstract concerns about the extensibility and robustness of the x-as-tile abstraction that I find hard to articulate. @trautmane if you have the same feeling, I'm happy to discuss this offline. Otherwise, I'd view this as an instance of the WET-principle and revisit this abstraction when (and if) we need to extend this a third time.

[minnerbe]

Are the first two rounds of optimization exactly the same (1000 iterations of 100% rigid with plateau width 250)? In this case, we could simply concatenate them (i.e., 2000 iterations of 100% rigid with plateau width 250).