Add metadata/descriptions for voice file variations

[EmZod]

Problem

When multiple variations of the same voice exist (e.g., morgan_freeman.wav, morgan_freeman2.wav, morgan_freeman3.wav), there's no way to know what differentiates them without trying each one.

Perspective: As an AI agent, when a user requests "Morgan Freeman voice," I see 3 files with no metadata. I can't make an informed choice - is variation 3 longer sample? Different tone? Better quality? Third take?

Current Behavior

$ ls ~/.chatter/voices/
morgan_freeman.wav    # ??? 
morgan_freeman2.wav   # ???
morgan_freeman3.wav   # ???
alan_watts.wav        # ???

No way to distinguish without trial-and-error.

Proposed Solutions

Option 1: Companion JSON metadata

~/.chatter/voices/
├── morgan_freeman.wav
├── morgan_freeman.json       # {"duration": "30s", "tone": "authoritative", "notes": "Best for narration"}
├── morgan_freeman2.wav
├── morgan_freeman2.json      # {"duration": "15s", "tone": "conversational", "notes": "Shorter sample"}

Option 2: Voice listing command

$ speak voices --list
morgan_freeman.wav       30s | Authoritative tone, best for narration
morgan_freeman2.wav      15s | Conversational, shorter sample  
morgan_freeman3.wav      52s | Longest sample, most expressive
alan_watts.wav           32s | Philosophical tone, slow cadence

Option 3: Descriptive filenames (breaking change)

morgan_freeman_narration_30s.wav
morgan_freeman_conversational_15s.wav
morgan_freeman_expressive_52s.wav

Recommended Solution

Option 2 (voice listing command) - No breaking changes, easily extensible, works with existing files.

Bonus: Combine with --preview flag for quick sampling:

speak voices --list --sample    # Plays 3-second sample of each voice

Impact

• Medium - Improves voice selection UX
• Reduces trial-and-error
• Enables agents to make informed choices