Support '0' value for parse_capacity_limit()

[mkleen]

Which issue does this PR close?

• None

Rationale for this change

This extends parse_capacity_limit() to support `0` to set a limit of 0 instead of using `0K`.
This simplifies configuration and avoids confusion with the word `OK` (Okay).
This is based on the suggestion from @martin-g.

Usage:

SET datafusion.runtime.example_limit = '0'

instead of:

SET datafusion.runtime.example_limit = '0K'

What changes are included in this PR?

see above.

Are these changes tested?

Yes.

Are there any user-facing changes?

Yes.

[kosiew]

@mkleen
The change looks good overall and I did not find any blocking issues. I left one small suggestion around SQL-level regression coverage to help ensure the config path stays wired correctly.

[kosiew]

Nice addition to cover the helper directly 👍

Since the intended user-facing flow is SET datafusion.runtime.* = '0', it could also be helpful to add a small SQL-level regression test or SLT that exercises something like SET datafusion.runtime.memory_limit = '0' and verifies SHOW returns the expected value.

That would help catch any future wiring regressions between the parser and the config layer.

[kosiew]

@mkleen

datafusion/execution/src/runtime_env.rs
in fn create_runtime_config_entries, the runtime config descriptions shown by information_schema/SHOW still say capacity limits support only K/M/G suffixes.

Since this PR adds bare 0 as a supported SQL-facing value for all parse_capacity_limit callers, consider updating the descriptions for memory_limit, max_temp_directory_size, metadata_cache_limit, and list_files_cache_limit to mention the special 0 value.