Which issue does this PR close?

• Closes panic on SET datafusion.runtime.* when value ends with non-ASCII byte (split_at char boundary) #22187.

Rationale for this change

Runtime SET datafusion.runtime.* parsing split the last byte off capacity and duration values with split_at(value.len() - 1). For values ending in a multi-byte UTF-8 character, that byte index can fall inside the final character and panic before DataFusion can return a planner error.

What changes are included in this PR?

• Adds a small UTF-8-aware helper that splits a config value at the start of its final character.
• Reuses that helper for deprecated memory limit parsing, current capacity limit parsing, and duration segment parsing.
• Adds unit coverage for non-ASCII invalid suffixes.
• Adds SQL runtime coverage for all reported reachable runtime keys.

Are these changes tested?

Yes.

• cargo test -p datafusion --lib test_parse_
• cargo test -p datafusion --test core_integration test_runtime_config_non_ascii_suffix_returns_error
• cargo fmt --check
• cargo clippy -p datafusion --lib --test core_integration -- -D warnings
• git diff --check

Are there any user-facing changes?

Yes. Malformed runtime config values ending in non-ASCII characters now return planner errors instead of panicking.

Scope note

The change is limited to avoiding byte-boundary panics while preserving the existing parser/error paths for malformed runtime config values.