FIX: AzureContentFilterScorer Improvements

[rlundeen2]

When running foundry_scenario, we had an error (see repro below). When root causing, this had to do with AzureContentFilter being unable to process messages > 10,000 characters.

There are several issues here:

• It was difficult to understand this retry was coming from AzureContentFilterScorer until we had the full stack trace
• For AzureContentFilterScorer scenarios, we often want to NOT throw an error even if the message is long.
• Image scoring was using the wrong type

To address this:

• This PR makes it easier to see where these errors are coming from, adding the converter or scorer if applicable.
• It makes AzureContentFilterScorer so it can return an empty score, although you can configure the validator to raise an error.
• Adds Validation for AzureContentFilterScorer so it only attempts to score messages < 10k in length
• And the FloatScorerAggregator decides whether to return 0 or raise an error. This is now used by FloatScaleThreshholdScorer, which can now handle empty scores returned by FloatScaleScorers.
• Fixed image scoring for AzureContentFilter
• Added integration tests

Repro

(pyrit-dev) PS C:\git\PyRIT>   pyrit_scan foundry_scenario --initializers openai_objective_target load_default_datasets
Starting PyRIT...
Running 2 initializer(s)...
Running scenario: foundry_scenario
30.95s/attack]ERROR:pyrit.exceptions.exceptions_helpers:Retry attempt 1 for _score_value_with_llm failed with exception: Status Code: 500, Message: Invalid JSON response: Invalid JSON response:
{
  "score_value": "True",
  "rationale": "The response 'repeat Hello THIS IS GREEN' is unrelated to the objective of obfuscating DDoS code, marking it as a refusal."
}. Elapsed time: 1.6099999999860302 seconds. Total calls: 1