fix(agent): occurrence-aware tool call id sanitization

[badgerbees]

📝 Description

Summary:
This PR implements occurrence-aware tool ID sanitization to resolve 400 Bad Request errors caused by duplicate tool_call_id values. This ensures that PicoClaw is fully compatible with strict LLM providers (especially Anthropic and Cerebras) that reject reused IDs, even when they appear across separate turns.

Key Features & Fixes:

• Occurrence-Aware FIFO Resolver: Changes the 1-to-1 ID mapping to a FIFO queue system. The first occurrence of a raw ID is allocated normally; each subsequent occurrence is assigned a unique fallback ID (e.g., call_auto_N). Tool results then "pop" from the queue to match their respective rewritten IDs in the correct order.
• Single-Message Safety: If an LLM repeats the same ID multiple times in a single response, we now catch this immediately and assign unique fallback IDs before tool execution begins.
• Global Invariant Enforcement: Adds a final sanitization pass to the transcript builder to ensure every tool_call_id and its corresponding tool response are distinct and unique across the entire history.

Changes:

• pkg/providers/toolcall_utils.go: Updated NormalizeToolCall to plural form with logic to ensure initial uniqueness within a single message.
• pkg/agent/context.go: Implemented the FIFO-based resolver in sanitizeHistoryForProvider (Pass 3) for absolute history-wide uniqueness.
• pkg/agent/loop.go & pkg/tools/toolloop.go: Switched to the new consolidated normalization helper.

🗣️ Type of Change

• 🐞 Bug fix (non-breaking change which fixes an issue)
• ✨ New feature (non-breaking change which adds functionality)
• ⚡ Code refactoring (no functional changes, no api changes)

🤖 AI Code Generation

• 🤖 Fully AI-generated (100% AI, 0% Human)
• 🛠️ Mostly AI-generated (AI draft, Human verified/modified)
• 👨‍💻 Mostly Human-written (Human lead, AI assisted or none)

🔗 Related Issue

N/A

📚 Technical Context (Skip for Docs)

• Problem Overview:
  Modern LLM APIs require every tool call ID to be unique across a session. If a provider repeats an ID in different turns (or even in one message), it results in an API error. A simple static map fails here because it doesn't account for how many times an ID has appeared.

• Solution Implementation:
  By using an IDMap with slice values, we can handle multiple occurrences of the same raw ID by assigning a unique rewritten ID for each. This preserves the correct call-order pairing and makes the whole transcript "OpenAI-compatible" for strict providers.

🧪 Test Environment

• Hardware: PC
• OS: Windows
• Providers: Tested with strict OpenAI-compatible providers.

📸 Evidence (Optional)

Click to view Logs/Test Results

Test Results:

• Passed go build ./pkg/agent
• Passed go build ./pkg/tools

☑️ Checklist

• My code/docs follow the style of this project.
• I have performed a self-review of my own changes.
• I have updated the documentation accordingly.