[Relax][TFLite] Add test coverage for Reduction operations (#18971)

[rknastenka]

Closes part of #18971

---

Description

This PR improves the test coverage of the TFLite frontend to Relax converter by adding comprehensive tests for the "Reductions" group.

Operations covered:

• SUM (tf.reduce_sum)
• MEAN (tf.reduce_mean)
• REDUCE_MAX (tf.reduce_max)
• REDUCE_MIN (tf.reduce_min)
• REDUCE_PROD (tf.reduce_prod)

Changes made:

• Added a parameterized testing function (test_reduction_ops) to cover combinations of the aforementioned reduction operators.
• Covered a variety of axis configurations, including positive scalars, lists of axes, negative indices, and None (global reduction).
• Tested with different combinations of the keepdims flag (True/False) and dtypes (float32/int32).
• Handled the representation of global reductions (axis=None) in the expected Relax IR Module _make_reduce_expected utility by expanding it to all input axes, perfectly mirroring the frontend's output graph structure (list(range(len(input_shape)))).

Testing

pytest tests/python/relax/test_frontend_tflite.py::test_reduction_ops

[gemini-code-assist]

Code Review

This pull request introduces a comprehensive test suite for TFLite reduction operations in the Relax frontend, covering sum, mean, max, min, and product operations. It includes a new helper function, _make_reduce_expected, to generate the expected Relax modules and utilizes parameterized testing to verify various input shapes, reduction axes, data types, and dimension retention settings. I have no feedback to provide.

[tlopex]

LGTM