refactor(rust/sedona-functions): centralize bounding functions

[paleolimbot]

This PR makes bounding functions for geography generic, implemented in sedona-functions using the bounder present in the config options. This lets geography bounding occur based on runtime injection of a bounder, rather than compile time features present in the sedona build.

I checked the bounding benchmarks for sedona-functions and didn't find any regressions.

Closes #1037.

[jiayuasu]

Could we register the bounder in new_from_context() too? With the sedona/s2geography feature enabled, this test now fails:

#[tokio::test]
async fn geography_bounds_default_context() {
    let ctx = sedona::context::SedonaContext::new();
    ctx.ctx
        .sql("SELECT ST_XMin(ST_GeogFromText('POINT (1 2)'))")
        .await.unwrap()
        .collect().await.unwrap();
}

The error is ST_XMin() requires a bounder for Spherical edges, which is not registered in this session. new_from_context(SessionContext::new()) has the same problem. new_local_interactive() works because it installs the bounder.

[jiayuasu]

Is geography support meant to survive a native UDF export/import roundtrip? I reproduced a failure at the Rust FFI boundary; the equivalent Python example is:

import sedonadb
from sedonadb.udf import sedona_native_scalar_udf

con = sedonadb.connect()
capsules = con.funcs.st_envelope.__sedonadb_scalar_udf__()
con.register(sedona_native_scalar_udf(capsules, name="rt_envelope"))

con.sql("SELECT ST_Envelope(ST_GeogFromText('POINT (1 2)'))").to_arrow_table()
con.sql("SELECT rt_envelope(ST_GeogFromText('POINT (1 2)'))").to_arrow_table()

The direct call works, but the imported call reports ST_Envelope() requires a bounder for Spherical edges, which is not registered in this session. The C wrapper invokes the exported kernel with None for ConfigOptions, so it cannot find the session's bounder. ST_XMin has the same issue; its previous S2 kernel worked through this roundtrip.