The `FoldEqZeroSTDistance` normalization rule normalizes expressions in
the form `st_distance(a, b) = 0` to `st_intersects(a, b)`. This
normalization rule is important because it allows an inverted index to
be used for more queries.

Previously, `FoldEqZeroSTDistance` did not handle the third argument of
`st_distance` for the geography overload, `use_spheroid`. The optimizer
panicked when a user supplied this third argument because it attempted
to find an `st_intersects` overload with the three arguments and no such
overload exists.

In addition to the panic, this rule was incorrectly firing in cases
where it shouldn't have.

From the [`ST_Distance` PostGIS docs](https://postgis.net/docs/ST_Distance.html):

> For geography types defaults to return the minimum geodesic distance
> between two geographies in meters, compute on the spheroid determined
> by the SRID. If use_spheroid is false, a faster spherical calculation
> is used.

From the [`ST_Intersects` PostGIS docs](https://postgis.net/docs/ST_Intersects.html):

> For geography, this function has a distance tolerance of about 0.00001
> meters and uses the sphere rather than spheroid calculation.

In summary, `st_distance` calculates on a spheroid by default, and only
on a sphere if `use_spheroid` is explicitly false. `st_intersects`
calculates on the sphere always. Therefore, this rule can only apply for
geography types if `use_spheroid=false` is explicitly passed.

This commit ensures that `FoldEqZeroSTDistance` only matches
`st_distance` functions with geography arguments if `use_spheroid=false`
is included in the function call. The panic is fixed by stripping the
`use_spheroid` argument from the list of arguments used to lookup the
`st_intersects` overload.

Fixes cockroachdb#67235

Release note (bug fix): Two bugs have been fixed which affected
geospatial queries with the `st_distance` function. The first caused
errors for filters of the form `st_distance(g1, g2, use_spheroid) = 0`.
The second could cause incorrect results in some cases. It incorrectly
transformed filters in the form `st_distance(g1, g2) = 0` when `g1` and
`g2` are geographies to `st_instersects(g1, g2)`. This is not a valid
transformation because `st_distance` makes spheroid-based calculations
by default while `st_intersects` only makes sphere-based calculations.