Feat: show_metrics() and stream_logs() helper functions

[ehsu3]

Issue #, if available: 'Monitoring training/eval jobs with CW logs' task

Description of changes:

• Added cloudwatch_metrics.py with CW log parsing for training_loss, lr, reward_score
• Supports SMTJ and SMHP
• Added show_metrics() on BaseTrainer with start_time/end_time, step range filtering
• Added stream_logs() on BaseTrainer (SMTJ auto-stops on completion, SMHP via Ctrl+C)
• show_metrics() generates separate subplots per metric
• Added pandas to pyproject.toml for generating graphs
• Added 18 unit tests covering parsing, fetching, end-to-end, and streaming
• The two functions can be used on a 'trainer' object that has a job name attached. If the session ends, the user can make a placeholder trainer object and add the job name so it's able to use the functions again.
• These functions only work training jobs since eval jobs don't have metrics stored in the CW logs and the logs don't have enough information for stream_logs() to provide information on progress. However, the latter can be added if we want.
• Added changes requested from initial PR (clean up code, use logger > print, notebook example, update parameters)

Tests:

• Ran both methods on some completed and in-progress jobs to make sure they're able to capture the metrics and logs properly.

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

[amazeAmazing]

Is this something we need to gate to the Nova models?
The log streaming should still work for OSS models, but Nova models metric parsing is Nova model specific

[ehsu3]

Yes, good catch! I'm not familiar with the OSS model logs but they'd likely be different from Nova. I'll update the logic there, thanks!