We've been accumulating quite a lot of environment variables that define behavior in addition to the CLI options and other explicit mechanisms in the codebase. @bertsky started documenting all of those in 2d85503 and this PR goes a step further to bundle all the functionality and documentation in one place.

In ocrd_utils.config, there is a singleton instance config of ocrd_utils.config.OcrdEnvConfig that should be used for registering environment variables and parsing them.

For example to define a new environemnt variable OCRD_LOGGING_CONFIG, we register it:

from pathlib import Path
from ocrd_utils import config
config.add('OCRD_LOGGING_CONFIG',
description="Path to logging configuration",
validator: lambda val: Path.exists(),
parser: lambda val: Path(val))

This environment variable can now be accessed as config.OCRD_LOGGING_CONFIG which will return a pathlib.Path because the parser. Since we're defining validator, this will fail as soon as config.OCRD_LOGGING_CONFIG is accessed, if the value is not an existing file and config.OCRD_LOGGING.CONFIG will return a Path.

From the description field, we generate the epilog to the ocrd --help output listing all the variables.

In short, whenever we need a new environment variable X:

• register it in ocrd_utils.config
• add it to the CLI epilog
• use it as config.X