Add TableClient.read_rows for non-transactional primary-key reads

[zinal]

Pull request type

• Feature

What is the current behavior?

The Table Service gRPC API already exposes ReadRows (read specified keys non-transactionally from a single table), and other language SDKs wrap it on TableClient. The Python SDK generated the protobuf stubs but did not provide a public client method, so callers had to use SELECT or read_table.

Issue Number: N/A

What is the new behavior?

Adds TableClient.read_rows (sync and async, plus async_read_rows on the sync client) as a session-less unary call, following the same pattern as bulk_upsert:

key_types = ydb.BulkUpsertColumns().add_column("id", ydb.PrimitiveType.Uint64)
result_set = driver.table_client.read_rows(
    "/local/users",
    [{"id": 1}, {"id": 2}],
    key_types,
    columns=("id", "name"),
)
for row in result_set.rows:
    print(row.id, row.name)

• Missing keys are omitted from the result rather than raising an error
• columns is optional; omit it to return every column
• Unit tests cover request construction, response wrapping, and both the sync __call__ and async_read_rows client paths
• Integration tests in tests/table/test_table_client.py and tests/aio/test_table_client.py exercise create/upsert/read/column projection/missing table, matching other TableClient methods

Other information

This is a public API extension only; existing methods are unchanged.

Checked locally: tox -e black, tox -e style, tox -e mypy, tox -e py -- ydb -v (223 passed), and sphinx-build -b html docs docs/_build/html -q (no new errors). Docker was not available in this environment, so the new integration tests were not executed against a live YDB.

  

[zinal]

Actual PR here: ydb-platform#898