Update OnlineObservation and Template Docstrings

[HamedHemati]

This PR update template docstring and OnlineObservation.

The changes in the OnlineObersvation class now enable training models with dynamic modules in online strategies. Model adaptation is removed for each sub-experience, and instead, model adaptation and optimizer reset happen only in the boundary sub-experiences. Related to issue #1197.

[AntonioCarta]

We cannot support dynamic models only in the boundary-aware setting. We should just do the model adaptation at every step in the boundary-free.

[HamedHemati]

Latest update:

• Added necessary attributes for model adaptation after each sub-exp.
• Re-added model adaptation for task-free streams.

[coveralls]

Pull Request Test Coverage Report for Build 4158245331

• 3 of 3 (100.0%) changed or added relevant lines in 1 file are covered.
• 3 unchanged lines in 2 files lost coverage.
• Overall coverage decreased (-0.01%) to 73.761%

---

Files with Coverage Reduction	New Missed Lines	%
avalanche/logging/text_logging.py	1	56.08%
avalanche/benchmarks/scenarios/classification_scenario.py	2	86.31%

Totals
Change from base Build 4134543518:	-0.01%
Covered Lines:	13859
Relevant Lines:	18789

---

💛 - Coveralls