Large forward maps make Julia crash

When trying to calibrate to LESbrary simulations, the forward map typically has over half a million elements, making it impossible to store a covariance matrix for EKI. How do you guys want to handle this? We could

1. Look into using sparse matrices.
2. Store fewer time steps in `observations`. This would make it harder to animate the solution downstream.
3. Add a `time_range` attribute to `ConcatenatedOutputMap` that specifies which time steps to track--defaults to all time steps.
4. Delete `OceanTurbulenceParameterEstimation.jl` and sip guava juice from Costco while brainstorming alternative research topics.

@glwagner @navidcy What do you guys think?

Start by reading the issue's discussion and inspecting ConcatenatedOutputMap, observations, and OceanTurbulenceParameterEstimation.jl. Compare the listed approaches for handling large forward maps, then confirm the maintainers' decision before implementing anything. Done means an agreed approach prevents the Julia crash while preserving the required downstream behavior.

Written by the indexing model from the issue text.