[BUG] the 2 parameter versions of pow and log have too large of an error

[revans2]

Describe the bug
The underlying cudf implementation of POW and LOG both show larger than wanted floating point differences with the spark CPU version when the optional parameter values are very large or very small. It is likely that a different algorithm is used to compute these values so when overflow happens we get different results.

Steps/Code to reproduce bug
The python integration tests

spark-rapids/integration_tests/src/main/python/arithmetic_ops_test.py

Line 355 in 6a7a0ef

def test_logarithm():

and

spark-rapids/integration_tests/src/main/python/arithmetic_ops_test.py

Line 382 in 6a7a0ef

def test_columnar_pow(data_gen):

are both marked as xfail because of this.

Expected behavior
The floating point error is withing the expected bounds.