Added tensorboard to driver.py for logging

[infophysics]

Added a few config parameters to allow the use of tensorboard:

base:
  run_number: '00'       # defaults to 00, but can be anything
  world_size: 1
  iterations: 100000
  seed: 0
  unwrap: false
  log_dir: /global/homes/n/ncarrara/dune/mlreco/pid_study/pid_solo/
  log_step: 1
  overwrite_log: true
  tensorboard: true     # defaults to false
  train:
    weight_prefix: /global/homes/n/ncarrara/dune/mlreco/pid_study/pid_solo/
    save_step: 1000
    optimizer:
      name: Adam
      lr: 0.001

The changes to driver.py are:

1. Adding in the SummaryWriter for writing log output to tensorboard format.

from torch.utils.tensorboard import SummaryWriter

2. Adding in run number and tensorboard options from config:

# check for run number
        if 'run_number' not in base:
            base['run_number'] = '00'

        # check for tensorboard
        if 'tensorboard' not in base:
            base['tensorboard'] = False

3. Added default options to initialize_base:

def initialize_base(self, seed, dtype='float32', world_size=0,
                        log_dir='logs', prefix_log=False, overwrite_log=False,
                        parent_path=None, iterations=None, epochs=None,
                        unwrap=False, rank=None, log_step=1, distributed=False,
                        split_output=False, train=None, verbosity='info',
                        run_number='00', tensorboard=False):

4. Saving run number and tensorboard option as driver members:

self.run_number = run_number
        self.tensorboard = tensorboard

5. Saving the log_path as a member to driver and initializing tensorboard:

# Initialize the log
        self.log_path = os.path.join(self.log_dir, log_name)
        self.logger = CSVWriter(self.log_path, overwrite=self.overwrite_log)

        # set up tensorboard
        if (self.tensorboard and self.main_process):
            time = datetime.now()
            now = f"{time.year}.{time.month}.{time.day}.{time.hour}:{time.minute}:{time.second}"
            self.tensorboard_dir = f'{self.log_path.replace(".csv", "/")}/{self.run_number}/{now}/'
            self.tensorboard = SummaryWriter(
                log_dir=self.tensorboard_dir
            )

6. Logging the log_dict losses and accuracies to tensorboard:

# Report to tensorboard
            if (self.tensorboard and self.main_process):
                for key, value in log_dict.items():
                    if ("loss" in key or "accuracy" in key):
                        self.tensorboard.add_scalar(key, value, iteration)

One can easily spin up tensorboard from a notebook using the following commands:

import nersc_tensorboard_helper
%load_ext tensorboard
%tensorboard --logdir <location_of_log_path> --port 0
nersc_tensorboard_helper.tb_address()

You will see something like the following:

![image](https://github.com/user-attachments/assets/25d283fa-7870-47cd-825a-6c0fbe8723a3)

Tensorboard allows the comparison of logged output over several runs, which is quite convenient.  This will also work for submitted jobs.

![image](https://github.com/user-attachments/assets/a94ed629-2690-4533-8394-b44ca0ede333)
![image](https://github.com/user-attachments/assets/1a51b277-0e87-450d-9e6c-e6804278e4ef)