Reduce CI flakiness in three iceberg test modules

[sfc-gh-mslot]

Problem

CI on `main` has been flapping over the past few weeks. Three failure
modes account for most of the noise:

1. `test_data_file_pruning.py::test_simple_data_pruning_for_data_types` —
   fails with `AssertionError: failed to refresh object catalog table`
   when the object-store catalog refresh hasn't observed the pushed
   metadata yet. The structurally identical test in
   `test_partition_pruning.py` already has `@pytest.mark.flaky(reruns=2)`
   (added in bbd1176, May 8); this sibling was missed.
2. `test_polaris_catalog_writable.py::test_sequences_serial_and_generated_columns[rest]` —
   fails with `AssertionError: Data file column stats mismatch` when
   the pg_catalog data-file stats and the REST metadata diverge by a
   row, which I've seen on multiple unrelated PRs and on `main`
   (26891987114, 26848484450, 26773102088, 26753357058, 27272760340).
3. `test_object_store_catalog.py::{test_schema_mismatch, test_partitioned_read_only, test_unsupported_modifications_for_read_only, test_multiple_readers_writers}` —
   fails with `psycopg2.errors.DuplicateSchema: schema "object_store_sc1" already exists`.
   This isn't really a flaky test — it's a cascade. When an earlier
   test in the module fails midway through (most recently due to a
   missing `pg_lake_ducklake` extension), it skips its
   `DROP SCHEMA ... CASCADE` and the next several tests fail on the
   leftover schema instead of running.

Solution

Three small, narrow changes:

• `test_simple_data_pruning_for_data_types`: add
  `@pytest.mark.flaky(reruns=2)` to match its sibling in
  `test_partition_pruning.py`.
• `test_sequences_serial_and_generated_columns`: add
  `@pytest.mark.flaky(reruns=2)`.
• `test_object_store_catalog.py`: add a module-scoped autouse fixture
  that runs `DROP SCHEMA IF EXISTS object_store_sc1, object_store_sc2 CASCADE`
  before each test. A single failure in this module no longer poisons
  the rest of it.

The first two are defensive masking and don't fix the underlying
race / consistency issue, but they match the prevailing convention in
the repo (`pytest-rerunfailures` was added in bbd1176 specifically for
this kind of test-harness-side flapping). The third is a real fix for
the cascade.

Test plan

• CI passes
• None of the three target tests appear in subsequent failure logs

[sfc-gh-okalaci]

test_data_file_pruning.py::test_simple_data_pruning_for_data_types —
fails with AssertionError: failed to refresh object catalog table
when the object-store catalog refresh hasn't observed the pushed
metadata yet. The structurally identical test in
test_partition_pruning.py already has @pytest.mark.flaky(reruns=2)
(added in bbd1176, May 8); this sibling was missed.

This was not missed, I think I objected to that, because I couldn't realize that it was due to object_store catalog.

It felt like pruning should be very deterministic, but then I think I'm fine with this explanation now.