fix: use ID-based lookup for parallel tool finalization

[daniel-lxs]

Summary

Replaces map-based index tracking with direct ID lookup when finalizing streaming tool calls. This prevents duplicate rendering of parallel tools by ensuring idempotent processing.

Problem

When parallel tools are called, they could sometimes be rendered twice because:

1. Multiple code paths can finalize the same tool (finalization loop, sweep fix, pending updates)
2. The streamingToolCallIndices map-based lookup could reference tools that were already processed

Solution

Use findIndex with partial === true check instead of streamingToolCallIndices.get(event.id):

• Only match blocks that are still partial === true
• If a tool was already finalized (partial = false), skip it
• Works for both sequential and parallel tool calling modes

Changes

• src/core/task/Task.ts: Replace map-based index lookup in finalization loop with ID-based lookup

Testing

All 40 related tests pass:

• 10 duplicate-tool-use-ids tests ✓
• 7 NativeToolCallParser tests ✓
• 23 presentAssistantMessage tests ✓

Risk Assessment

Low risk - The change is targeted and defensive:

• Only affects the tool finalization code path
• Idempotent by design (skips already-finalized tools)
• No changes to tool execution logic

[ghost]

Rooviewer   See task on Roo Cloud

Review complete. No issues found.

The change is clean and addresses the duplicate rendering problem for parallel tools effectively. The ID-based lookup with partial === true check ensures idempotent processing.

_{Mention @roomote in a comment to request specific changes to this pull request or fix all unresolved issues.}