[BUG]: Terminal streaming renderer corrupts output - characters doubled/truncated during streaming

## Description

The Copilot CLI terminal renderer corrupts streamed output, producing doubled characters, truncated tokens, and repeated lines. This affects both the thinking/reasoning phase output and the final assistant response output.

## Steps to Reproduce

1. Ask Copilot CLI any question that produces a moderately long response
2. Observe the output as it streams to the terminal

## Observed Behavior

Output is visibly corrupted during streaming. Examples seen:
- Partial tokens rendered before full token is available (e.g. `apil` instead of `apiVersion`)
- Characters doubled mid-word (e.g. `obserrvability`, `permpermissions`)
- Words repeated mid-sentence (e.g. `New ReliNew Relic`, `useruser`)
- Lines repeated (e.g. `metadata:` block appearing twice)
- Thinking/reasoning tokens streamed visibly with same corruption artifacts

Example corrupted output observed:
```
apil
apiVersion: rbac.authorization.k8s.io/v1
...
- apiGroups: [piGroups: [""]
```

## Expected Behavior

Output should render cleanly with no corruption, doubled characters, or repeated lines.

## Environment

- Copilot CLI version: 1.0.61
- OS: Darwin (macOS)

## Notes

The underlying content being generated is correct — the issue is purely in how the CLI streams and renders tokens to the terminal.

Inizia riproducendo la corruzione nel renderer del terminale di Copilot CLI su macOS, confrontando lo streaming di thinking/reasoning e quello della risposta finale. Traccia il modo in cui il renderer gestisce i token parziali e gli aggiornamenti delle righe; il lavoro è completato quando l'output trasmesso corrisponde al contenuto sottostante, senza caratteri raddoppiati, troncamenti o righe ripetute.

Scritto dal modello di indicizzazione a partire dal testo della issue.