FIX: preserve rationale-less float scores

[Barry (barry166)]

Description

Fixes #2880.

Float-scale aggregate rationales now include every constituent score, including scores without a rationale. This keeps the output consistent with the true/false and undetermined aggregation paths and preserves which scores contributed to the result.

Tests and Documentation

• uv run --offline pytest -q tests/unit/score/test_float_scale_score_aggregator.py (39 passed)
• uv run --offline ruff check pyrit/score/float_scale/float_scale_score_aggregator.py tests/unit/score/test_float_scale_score_aggregator.py
• git diff --check

[Roman Lutz (romanlutz)]

The new lines still don't identify which Azure harm category produced each value. For example, rationale-less AzureContentFilterScorer scores for Hate=0.2 and Violence=0.8 produce AzureContentFilterScorer 0.2: and AzureContentFilterScorer 0.8:. The aggregate lists both categories, but cannot tell a reader which category caused the 0.8 MAX or a threshold verdict. Could we include each score's category in its line (ideally in the shared formatter) and test this multi-category case? That would make the newly preserved entries useful for tracing the result.