… / getOption

DataFusion's SessionConfig carries roughly 200 keys split across seven sections (datafusion.execution.*, datafusion.optimizer.*, etc). The Java SessionContextBuilder introduced in apache#28 covers six of them with named setters, and there is no Java surface to read any config value back at all. Rather than ship ~200 named get/set pairs one at a time, mirror DataFusion's existing string-keyed surface (ConfigOptions::set) as a generic escape hatch on the builder + context.

Adds two additions to session_options.proto: a `repeated ConfigOption options = 7;` field plus the matching ConfigOption message. `repeated` is used instead of `map<string,string>` because protobuf maps decode into a Rust HashMap whose iteration order is randomized -- that would silently break overlapping-key cases like `datafusion.optimizer.enable_dynamic_filter_pushdown` (whose setter has side effects on the per-operator `enable_*_dynamic_filter_pushdown` flags). The Java builder gains setOption(key, value) and setOptions(Map). Java-side storage is a LinkedHashMap so caller insertion order is preserved end to end.

On the native side, free-form options are applied via config.options_mut().set(k, v)? before SessionContext construction. Map entries are applied after the typed setters so an explicit setOption call wins over a typed setter for the same knob, and within the entry list the caller's last write wins -- both for same-key duplicates (LinkedHashMap dedups) and for overlapping side-effect keys. The ? form (rather than SessionConfig::set_str's .unwrap()) means unknown keys or unparseable values surface as a RuntimeException with DataFusion's error message instead of panicking the JVM.

Adds SessionContext.getOption(key) on the constructed context (not on the builder, since the value reflects "what DataFusion actually compiled" -- only knowable post-construction). The native side walks ctx.copied_config().options().entries() and returns ConfigEntry.value as a String, or null if the key is known but unset and has no default. Unknown keys throw RuntimeException to mirror setOption's strictness.

datafusion.runtime.* keys (memory limit, temp directory, cache sizes) live on a separate RuntimeEnv config object and have several upstream-shaped round-trip pitfalls that don't apply to the SessionConfig subtree (lazy default tempdir, per-session datafusion-Xxxxxx spill suffixes, OS-specific path separators, integer K/M/G truncation, sub-1KB byte formatting, the unlimited sentinel, multi-statement SET clobbering). Both setOption and getOption reject runtime keys with a clear error pointing at the typed memoryLimit() / tempDirectory() setters; round-trippable runtime support is tracked as a follow-up PR that needs a per-context side-cache.

Tests cover the proto round-trip with explicit on-the-wire ordering assertions, bulk setOptions, null rejection, override-typed-setter semantics (asserted by reading the value back via getOption), last-write-wins for repeated keys, an unknown-key error path on both set and get, default-fallback on get, a closed-context guard on get, the runtime-key rejection on both set and get with messages that point at the typed setters, and the order-preservation case where setting the umbrella `enable_dynamic_filter_pushdown` flag followed by an explicit `enable_topk_dynamic_filter_pushdown=false` correctly leaves topk disabled (the override winning over the umbrella's side effect).

Common knobs this unlocks include parquet pushdown_filters / bloom_filter_on_read, optimizer prefer_hash_join, execution time_zone, sql_parser dialect, and explain show_statistics -- previously inaccessible from Java without adding a named get/set per key.