Increase maximum module files payload using protobuf streams

Problem

Coder has a 4MB protobuf message size limit that prevents templates with large Terraform modules from being processed. When dynamic parameters require module files from ~/.terraform/modules that exceed this limit, template creation fails.

Solution

This PR implements protobuf streaming to handle large module files by:

1. Streaming large payloads: When module files exceed the 4MB limit, they're streamed in chunks using a new UploadFile RPC method
2. Database storage: Streamed files are stored in the database and referenced by hash for deduplication
3. Backward compatibility: Small module files continue using the existing direct payload method

Key Changes

• New protobuf streams: Added UploadFile RPC and DataUpload/ChunkPiece message types
• Smart fallback: Automatically detects oversized payloads and switches to streaming
• File deduplication: Module files are deduplicated per template version using content hashes
• Testing: Added test coverage for the streaming upload functionality.

Architecture

• Coderd ↔️ Provisionerd: Uses new streaming RPC when payload exceeds limits.
  

  coder/provisionerd/proto/provisionerd.proto

  Lines 220 to 222 in e880894

  	// UploadFile streams files to be inserted into the database.
  	// The file upload_type should be used to determine how to handle the file.
  	rpc UploadFile(stream UploadFileRequest) returns (Empty);

• Provisionerd ↔️ Provisioner: Extends existing stream with new message types for in-memory communication

This change enables templates with arbitrarily large Terraform modules while maintaining performance through deduplication and preserving the existing fast path for smaller payloads.

TODO:

• Bump protobuf versions
• Implement backwards compatible?
  • Do not upload large files to old coderds
• Verify forward compatible
• Add e2e test with a large module

Things that I wish I could change

• We have 2 proto files, and they have overlapping types. Ideally DataUpload and Chunk would be defined in 1 place and have their types reusable in both contexts 😢.