Support dumping training logs for TensorBoard visualization toolkit.

[classicsong]

Issue #, if available:
#988

Description of changes:
Add a TensorBoard tracker that will save training loss, validation scores and test scores into TensorBoard logs.

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[thvasilo]

To support optional dependencies we do the following @classicsong

Pandas has their own pandas.compat._optional.import_optional_dependency We can use a similar mechanism.
See https://pandas.pydata.org/pandas-docs/version/1.4/development/contributing_codebase.html#optional-dependencies and https://pandas.pydata.org/pandas-docs/version/1.4/getting_started/install.html#install-optional-dependencies

Optional imports are done inline, using something like

def generate_numba_apply_func(
    func, nogil=True, nopython=True, parallel=False
) -> Callable[[npt.NDArray, Index, Index], dict[int, Any]]:
    numba = import_optional_dependency("numba")
    jitted_udf = numba.extending.register_jitable(func)

So in this case we could do

class GSTensorBoardTracker(GSSageMakerTaskTracker):
    def __init__(self, log_report_frequency, log_dir=None):
        super().__init__(log_report_frequency, log_dir)
        try:
            tensorboard = importlib.import_module("torch.utils.tensorboard")
        except ImportError as err:
            msg =  (
                "GSTensorBoardTracker requires tensorboard to run. "
                "Please install the tensorboard Python package.")
            raise ImportError(msg) from err
        
        self._writer = tensorboard.SummaryWriter(log_dir)

This ensures we only try to pull the dependency if we try to instantiate a GSTensorBoardTracker object and not
at module import time.

Then we can modify our setup.py to include tensorboard as an extra dependency:

setup(
    # Metadata
    name='graphstorm',
    version=VERSION,
    python_requires='>=3.8',
    description='GraphStorm',
    long_description_content_type='text/markdown',
    license='Apache-2.0',

    # Package info
    packages=find_packages(where="python", exclude=(
        'tests',
    )),
    package_dir={"": "python"},
    package_data={'': [os.path.join('datasets', 'dataset_checksums', '*.txt')]},
    zip_safe=True,
    include_package_data=True,
    install_requires=requirements,
    ext_modules=extensions,
    cmdclass=cmdclass,
    
    extras_require={
        'visualization': [
            'tensorboard', # Add minimum version if needed, e.g. 'tensorboard>=2.11.2'
        ]
    },
)

then when users want tensorboard visualization they can install pip install graphstorm[visualization] to pull in the extra tensorboard dependency.