Add weighted voice blending support to combine endpoint

[RaphaelRemond]

Add weighted voice blending support to combine endpoint

This update introduces the ability to assign custom weights to each voice when generating a combined voice tensor. Previously, all voices were implicitly treated as having equal weight, limiting control over the blending process.

The new implementation:

• Parses voice entries with optional weight values (e.g. "emma:0.7+isabella:0.3")
• Validates and normalizes weights to ensure consistent mixing
• Loads tensors efficiently and applies weighted blending
• Produces a properly named output file reflecting the applied ratios

This change enables more flexible and accurate voice composition.