Bug fix: schema evolution fix-to-var reads followup.

[bekadavis9]

Schema evolution bug fix: Reads no longer fail after dropping a fixed attribute and adding it back as var-sized.

Followup to #5321 to address a previously-missed case caught by TileDB-Inc/TileDB-Py/#2083

[sc-55085]

---

TYPE: BUG
DESC: Schema evolution bug fix: Reads no longer fail after dropping a fixed attribute and adding it back as var-sized, part 2.

[kounelisagis]

Just to make sure we don’t lose or forget (talked with @bekadavis9 about it).

If I change my test to only have a single argument:

def test_schema_evolution_drop_fixed_attribute_and_add_back_as_var_sized(tmp_path):
    ctx = tiledb.default_ctx()
    uri = str(tmp_path)
    attrs = [
        tiledb.Attr(name="a", dtype=np.int32),
    ]
    dims = [tiledb.Dim(domain=(1, 10), dtype=np.int32)]
    domain = tiledb.Domain(*dims)
    schema = tiledb.ArraySchema(domain=domain, attrs=attrs, sparse=False)
    tiledb.Array.create(uri, schema)

    data = np.arange(1, 11)
    with tiledb.open(uri, "w") as A:
        A[:] = {"a": data}

    se = tiledb.ArraySchemaEvolution(ctx)
    se.drop_attribute("a")
    se.array_evolve(uri)

    se = tiledb.ArraySchemaEvolution(ctx)
    newattr = tiledb.Attr("a", dtype="S", var=True)
    se.add_attribute(newattr)
    se.array_evolve(uri)

I am getting tiledb.cc.TileDBError: ArraySchema: Array schema check failed; No attributes provided coming from the last line.
I think this is not ideal behavior. I understand that ArraySchema should always have an argument, so I believe an error should be thrown when the attribute is dropped instead(?).

[kounelisagis]

After updating the test in TileDB-Inc/TileDB-Py#2083, it seems we now have the expected behavior. After discussing with @bekadavis9, although the test I mentioned in my previous comment reveals a non-ideal behavior, addressing it would require a major rewrite of the SchemaEvolution class and is beyond the scope of this PR.

With that in mind, this looks good to me. Nice work!