fix(tool-guardrails): detect content-repetition loops across varying args (#60084)

[kyssta-exe]

Problem

The tool-loop guardrails in agent/tool_guardrails.py miss two loop shapes:

1. Varying-args / fixed-result loops: A tool call succeeds with different arguments on every call but keeps returning the same blocked/empty/error-page body. The existing _no_progress tracker is keyed by ToolCallSignature (tool name + args hash), so varying args resets the counter.
2. Repeated multimodal results: A vision tool result embeds a unique base64 payload every call, making str(result) unique even when the meaningful content is identical.

Fix

• Adds _semantic_result_hash() that strips base64 multimodal payloads before hashing, so vision-tool image reloads are detected
• Adds _strip_multimodal_content() that recursively replaces large base64 strings and _multimodal=True content blocks with placeholder markers
• Adds a _content_repetition tracker in ToolCallGuardrailController that maps (tool_name, semantic_hash) -> repeat_count, catching same-content results regardless of argument variation
• Tracking fires for ALL tools (not just idempotent ones), emitting a repeated_content_warning after the configured threshold

Fixes #60084

[alt-glitch]

This was generated by AI during triage.

Duplicate of #60087 — same content-hash-only result-repetition axis for #60084 in agent/tool_guardrails.py (both strip multimodal payloads before hashing and track (tool, semantic_hash) -> count), differing only in tracker/knob names. #60087 is the earliest-open version. Siblings #60148/#60094 already deduped into #60087.

[kyssta-exe]

Stale — 6-7 days without merge activity. Can resubmit if still needed.