FIX: CLI Usability Updates

[rlundeen2]

Summary

A set of CLI usability and diagnostic improvements for scenario runs (pyrit_scan), making progress reporting honest and errors actionable.

Changes

• Accurate technique progress count. ScenarioResult.get_techniques_used() now aggregates atomic-attack cells by display group, so the progress bar and summary show real technique counts (e.g. 2/14) instead of atomic-attack-cell counts that could exceed the total.
• Fix adversarial-chat resolution for simulated-conversation techniques. New AttackTechniqueFactory.resolve_adversarial_chat() lazily resolves the default adversarial target when a technique's seed group carries a simulated conversation, fixing adversarial_chat is required when seed_group has a simulated conversation config failures for techniques like many_shot_*.
• More actionable errors. Atomic-attack execution errors now include the atomic-attack name.
• Surface failures and retry pressure in the summary even on non-failing runs: per-failed-attack error type/message and total retry counts. Dropped the redundant Completed line (it always equalled Total Attacks).
• Stream retry warnings during polling. Each attack's retries are printed once as its result lands, including the component role/name and endpoint captured on the RetryEvents.
• --start-server output. Prints the log path on success and tails the log on crash/timeout.

Testing

• pre-commit clean (ruff format/check, ty).
• Affected unit suites pass (342 tests): CLI output/scan/server-launcher, scenario-run service, scenario result, technique factory, matrix builder.