docs: explain context growth in long agent runs

[cupkk]

Closes #2566.

This adds a short section to the memory tutorial that:

• shows why replaying accumulated memory makes total input-token usage grow quadratically with the step count;
• demonstrates how to inspect reported input-token usage; and
• uses an existing step callback to compact older tool observations.

The callback example also calls out its tradeoff: observation compaction is lossy and does not remove the rest of each historical step.

Validation:

• parsed the new Python example with ast;
• executed it against the repository's ActionStep and CallbackRegistry classes;
• checked Markdown fence balance and the numeric example in the formula.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.