FEAT: SeedSimulatedConversation to generate simulated conversations in attacks

[rlundeen2]

Refactor Seed Models, allowing Seeds to configure simulating conversation in Attacks.

• Separated SeedGroup and SeedAttackGroup into distinct files, where SeedAttackGroup is a subclass - specifically a SeedGroup with exactly one objective. I imagine in the future there will be more groups (e.g. SeedBenchmarkGroup).
• Added SeedSimulatedConversation to be able to configure simulating conversations in datasets. In the future I expect there will be more seed types (e.g. SeedExpectedResponse that might have an expected return value and required for benchmarks)
• Added code in AttackExecutor and Attack's FromSeedGroups to generate simulated conversations based on SeedSimulatedConversation

Multiple small refactors to clearly separate responsibilities along the way. In the future

To document, I added examples in 2_seed_programming that shows how Seeds translate to attack parameters, and included the new SeedSimulatedConversation

DB Migration Notes

The following failed before migration:

pytest .\tests\integration\memory\test_azure_sql_memory_integration.py::test_get_seeds_with_metadata_filter

Migration in test and prod:

ALTER TABLE [SeedPromptEntries] ADD [seed_type] NVARCHAR(50) NULL;

Then the test passes