Why?

This PR implements field-level optimization attributes for Rust, enabling users to control serialization behavior at the field level. This reduces serialization overhead by:

1. Using compact field ID encoding instead of field name encoding
2. Allowing explicit control over nullable and reference tracking flags
3. Supporting field skipping for transient data

What does this PR do?

New Field Attributes

Adds support for #[fory(...)] field attributes with the following options:

Usage Examples

#[derive(ForyObject)]
struct User {
    #[fory(id = 0)]  // Use compact field ID encoding
    name: String,
    
    #[fory(id = 1, nullable)]
    email: Option<String>,
    
    #[fory(ref = false)]  // Disable ref tracking for this Rc
    data: Rc<Data>,
    
    #[fory(skip)]  // Exclude from serialization
    cache: HashMap<String, String>,
}

Compact Field ID Encoding

Per xlang_serialization_spec.md, field IDs use TAG_ID encoding:

| 2 bits encoding (0b11) | 4 bits field_id | 1 bit nullable | 1 bit ref_tracking |

This provides ~28% smaller payload compared to field name encoding for structs with explicit field IDs.

Implementation Details

• field_meta.rs: New module for parsing #[fory(...)] attributes
• type_meta.rs: Updated FieldInfo with new_with_id() constructor and compact encoding
• misc.rs: Generates field metadata with nullable/ref_tracking/field_id
• util.rs: Updated fingerprint computation to include field metadata
• Merged with tuple struct support from main branch

Related issues

Closes #3004
#1017

Does this PR introduce any user-facing change?

• Does this PR introduce any public API change?
  • Adds optional #[fory(...)] field attributes
  • Fully backward compatible (no attributes = existing behavior)
• Does this PR introduce any binary protocol compatibility change?
  • Uses existing TAG_ID encoding from xlang spec

Benchmark

Field ID encoding produces ~28% smaller payloads:

• With field IDs: 98 bytes
• With field names: 137 bytes