feat: simplify attack logging with nested attempt_history

[kyuz0]

Dynamic attacks now keep one <id>-attack result with their representative input/response, overall verdict, and total attempt count. Supporting single-turn attacks attach intermediate candidates as nested attempt_history, simplifying logging and keeping result statistics tied to the attack's reported count.

• Collect history explicitly inside supporting attack modules, including Best-of-N, both LLM jailbreakers, and the workspace sample attack. Keep the existing four-element return tuple and legacy module signatures.
• Enable collection by default, with SPIKEE_ATTACK_HISTORY=false available to reduce result size. Remove the CLI flag, viewer checkbox, extra attack parameter, and alternative list return.
• Restore Crescendo, GOAT, and Echo Chamber's original conversation logging; remove duplicate candidate logging from multi_turn.
• Combine supplied candidate histories across repeated invocations and display them within the result detail. Preserve target responses when judging fails, and keep valid outcomes/counts when optional history is malformed.
• Continue reading legacy expanded result files. Update testing, custom-attack, and results documentation and the workspace environment example.

Validation on the rebased commit:

• All 479 functional tests passed on Python 3.11 with isolated package installation, covering legacy modules/results, counts, resume, extraction, rejudge, error paths, and history enabled/disabled.
• Ruff 0.16.6 lint and formatting checks passed for spikee/ and the three changed/new test files.
• Viewer checks passed on desktop and mobile.