c/sedona-s2geography: Reduce precision may have differing behaviour when some components collapse in a geography

I removed a test so that I can investigate properly from https://github.com/apache/sedona-db/pull/816 for ST_ReducePrecision for geography (which bigquery calls ST_SnapToGrid). We might also want to investigate naming them the same thing (but also investigate the specifics of the difference between ST_SnapToGrid and ST_ReducePrecision in PostGIS...I think snaptogrid is faster but maybe doesn't do validity checking in the same way).

```python
# Linestring: component collapses because the endpoints snap together
pytest.param(
"LINESTRING (0.01 0.02, 0.03 0.04)",
1.0,
"LINESTRING EMPTY",
id="linestring_collapse",
),
```

BigQuery gives `POINT (0 0)` here, which keep the collapsed component

Start with PR 816 and the removed pytest case for ST_ReducePrecision on geography, especially the linestring whose endpoints collapse. Compare the observed behavior with BigQuery's ST_SnapToGrid and the corresponding PostGIS functions, including validity handling. Done means the differing collapsed-component behavior is understood, agreed, and covered by an appropriate regression test.

Written by the indexing model from the issue text.