Optimized Embeddings CPU Usage

[satp42]

Added new settings to control embedding performance in packages/types/src/config.types.ts. Specifically:

1. embedding_batch_size (number, default: 64)
2. embedding_max_threads (number, default: 4)
3. embedding_max_connections (number, default: 8)

Modified packages/backend-server/src/main.rs to accept additional command-line arguments for batch size, max threads, and max connections.

Updated packages/backend-server/src/server/mod.rs:

• Added max_connections field to LocalAIServer
• Implemented a semaphore or connection counter to limit concurrent client connections (currently unlimited thread spawning)
• Configured rayon global thread pool using rayon::ThreadPoolBuilder before starting the server

Modified packages/backend-server/src/embeddings/model.rs:

• Added batch_size field to EmbeddingModel struct
• Replaced hardcoded batch size Some(1) at line 71 with configurable self.batch_size

Passed Configuration from Electron Main Process

Implemented Lazy Embeddings for Large Document Types

• Extended lazy embeddings logic to include ResourceTextContentType::PDF, ResourceTextContentType::Document, and ResourceTextContentType::Article
• These document types will get a generateLazyEmbeddings tag instead of immediate embedding generation
• Embeddings will then be generated on-demand when documents are accessed in chat/search

Optimized Chunking Strategy

• Increased max_chunk_size from 2000 to 2500 characters (reduces total chunks by ~20% while maintaining quality)
• Kept overlap_sentences at 1 for continuity
• This change reduced the number of embeddings needed per document

The expected impact of this PR:

• Batch size increase (1 → 64): reduction in CPU overhead due to better model utilization
• Thread pool limits: Prevents CPU saturation, keeps usage under control
• Connection limits: Prevents thread explosion during bulk uploads
• Lazy embeddings for large docs: Defers expensive operations until needed
• Larger chunks (2000 → 2500): fewer embeddings to generate and store

Related to #28