Investigate slimming down the container image

[jl-0]

Context

The Containerfile currently clones emit-sds-l2b and installs heavy conda dependencies (gdal, pandas, scipy, matplotlib) via pixi to support the full L2B post-processing pipeline (aggregator.py, group_aggregator.py, output_conversion.py).

However, tetracorder-lite's core scope is:

1. Run tetracorder (tetrapy run/setup/tetrun) — shells out to the compiled Fortran binaries
2. Build convolved libraries (tetrapy convolve) — pure Python, only needs numpy + click

Neither of these depends on gdal, pandas, scipy, matplotlib, emit-utils, or the emit-sds-l2b repo.

Proposal

Investigate whether the L2B dependencies can be removed (or made optional) to reduce image size and build time. Options:

• Remove entirely if full L2B processing doesn't belong in this container
• Multi-stage or separate image if both use cases need to coexist
• Optional pixi environment so the base image stays lean

Questions to resolve

• Is there a use case for running aggregator.py inside this container, or does that happen elsewhere in the pipeline?
• Does James's tetrapy patch command (which processes the expert file for L2B compatibility) need any of these deps?
• Could emit-sds-l2b be a proper pip-installable package instead of a raw git clone?