Added comprehensive tracking and cost calculation for OpenAI TTS models (tts-1, tts-1-hd).

What Changed

• Character-based billing: TTS models charge per character, not tokens - now we track that correctly
• Automatic detection: Works with existing track_openai() - no code changes needed
• Real costs: Shows actual OpenAI pricing ($0.000015/char for tts-1, $0.000030/char for tts-1-hd)
• Full metadata capture: Tracks model, voice, response format, speed, and all API parameters

API Coverage

• client.audio.speech.create() - Regular TTS calls
• client.audio.speech.with_streaming_response.create() - Streaming TTS calls
• All existing OpenAI tracking setups continue working

Technical Implementation

Backend Changes:

• Extended ModelPrice with inputCharacterPrice field
• Added character-based cost calculation in SpanCostCalculator
• Auto-detection of billing type (token vs character) in CostService
• Support for input_characters usage tracking

Frontend Updates:

• Extended UsageData interface for character usage
• Updated trace helpers to display character metrics
• Backward compatible with existing token displays

Python SDK:

• New OpenaiAudioSpeechTrackDecorator for TTS tracking
• OpenaiAudioSpeechUsage class for character-based usage
• Updated OpikUsageFactory to support TTS models
• Seamless integration with existing OpenAI tracking

Usage Example

import openai
from opik.integrations.openai import track_openai

client = track_openai(openai.OpenAI(), project_name="tts-demo")

# Automatically tracked with character usage and costs
response = client.audio.speech.create(
    model="tts-1",
    voice="alloy",
    input="This text will be tracked with accurate character-based billing."
)

Testing & Validation

✅ 22 comprehensive tests added

• 9 integration tests (end-to-end TTS tracking)
• 13 unit tests (character calculation, edge cases)

✅ Manual testing completed

• Real OpenAI API call validation
• Cost calculation accuracy verified
• Both regular and streaming API tested

✅ Zero breaking changes

• Fully backward compatible
• Existing projects work unchanged
• Optional character usage fields

Performance & Security

• No performance impact: Leverages existing infrastructure
• Minimal overhead: Simple character counting
• Secure: Uses existing API tracking patterns
• No new dependencies: Built on current architecture

/claim #2202 fixes #2202