Split Beam Python base image into a 'slim' version and 'regular/extended' version #20605
Description
Beam container image includes several dependencies that are not essential to Beam, but may be commonly used by some users. Some dependencies may be relevant only for particular runners.
Installing dependencies in Beam containers increases image size. For example, installing TensorFlow adds ~1G or so of space into the unpacked image.
Some Beam users may be size conscious, and would prefer an image with minimal dependencies.
Some Beam users and Beam runner maintainers may prefer an image that includes additional dependencies. These extended images may provide better user experience in some scenarios, but the overall image would be larger.
We should consider splitting beam image into slim and regular version to accommodate these different usage scenarios.
Imported from Jira BEAM-11166. Original Jira may contain additional context.
Reported by: tvalentyn.