Evaluate confirmation and transaction semantics

[rlippmann]

Evaluate whether Context Compiler should ever own confirmation or staged transition workflows.

Distinguish:

• host-owned confirmation workflows (for example rm -i style behavior)
• compiler-owned transactions (for example database autocommit-off style behavior)

Questions to resolve:

• Does the compiler need staged state transitions?
• Would commit/rollback be a better abstraction than pending clarification?
• When should interaction state become compiler-owned?

This is not a request to restore checkpoint/pending behavior; it tracks a future architectural decision.