FIX: Curate Scam scenario DEFAULT strategy

[varunj-msft]

Description

Part of standardizing our scenarios.

Previously the Scam scenario had default_strategy=ScamStrategy.ALL, so a run with no --strategies executed every technique, including the multi-turn PersuasiveRedTeamingAttack. That multi-turn attack runs several turns per objective and makes the default run slow and expensive, which isn't a good out-of-the-box default.

This change adds a curated DEFAULT aggregate and points the scenario's default at it:

• Adds DEFAULT = ("default", {"default"}) to ScamStrategy and tags the two single-turn techniques (ContextCompliance, RolePlay) with "default". PersuasiveRedTeamingAttack (multi-turn) is intentionally left out of DEFAULT.
• Flips default_strategy from ALL to DEFAULT.

So a plain pyrit_scan airt.scam now runs the two fast single-turn techniques. The multi-turn attack is still fully available via --strategies all, --strategies multi_turn, or --strategies persuasive_rta.

This follows the same manual-enum DEFAULT convention already used by the garak WebInjection and Doctor scenarios.

VERSION is bumped 1 -> 2 since the default run behavior changes.

Tests and Documentation

Tests (tests/unit/scenario/airt/test_scam.py):

• Added a strategy-enum test class asserting DEFAULT expands to exactly {context_compliance, role_play}, DEFAULT excludes persuasive_rta, and ALL still includes persuasive_rta.
• Added an end-to-end test that a no-strategy (default) run yields only the two single-turn atomic attacks.
• Made the existing "all" generation test pass ScamStrategy.ALL explicitly and assert all three techniques run.
• Updated the default_strategy and VERSION assertions.
  All scam unit tests pass; ran ruff, ruff format, and ty clean.

Documentation:

• Updated the Scam section of doc/scanner/airt.py (and the paired airt.ipynb) to list the DEFAULT aggregate and explain that the multi-turn technique is opt-in. The code sample uses an explicit strategy and is unchanged, so no re-execution was required; airt.ipynb was kept in sync with the .py.