[Feature][transform-v2] Vector Functions Support in SQL Transform

### Search before asking

- [x] I had searched in the [feature](https://github.com/apache/seatunnel/issues?q=is%3Aissue+label%3A%22Feature%22) and found no similar feature requirement.

### Description

Add comprehensive vector processing capabilities to SeaTunnel's SQL Transform engine, enabling users to perform vector operations directly within SQL queries for machine learning and AI workloads.

### Core Vector Functions

1. **VECTOR_REDUCE** - Vector dimensionality reduction with multiple algorithms
2. **VECTOR_NORMALIZE** - Normalize vector to unit length

### Supported Reduction Methods

- **TRUNCATE** - Simple truncation: keep first N dimensions
- **RANDOM_PROJECTION** - Gaussian random projection
- **SPARSE_RANDOM_PROJECTION** - Sparse random projection for memory efficiency

### Usage Scenario

_No response_

### Related issues

_No response_

### Are you willing to submit a PR?

- [x] Yes I am willing to submit a PR!

### Code of Conduct

- [x] I agree to follow this project's [Code of Conduct](https://www.apache.org/foundation/policies/conduct)

Start by locating SeaTunnel's SQL Transform engine and its existing function registration and vector-related handling. Review how transform functions are defined and tested, then define support for VECTOR_REDUCE with the listed reduction methods and VECTOR_NORMALIZE; done means these operations work directly in SQL queries with coverage for each method.

Written by the indexing model from the issue text.