feat(4): support streaming typed clients

[0xpolarzero]

Warning

This PR is part of a stacked PR chain.

Upstream merge PRs:
wevm/incur#144 ->
wevm/incur#145 ->
wevm/incur#143 ->
wevm/incur#146 ->
wevm/incur#147

Correctly based review stack in this fork:
#3 ->
#4 ->
#5 ->
#6 ->
#2

Overview

Adds async-generator streaming support to generated TypeScript clients and the existing structured RPC route.

API

// Commands declared with `async *run` return typed async iterables.
const stream = await client('project logs')({
  args: { id: 'p1' },
})
//    ^? AsyncIterable<{ line: string }>

for await (const chunk of stream) {
  //             ^? { line: string }
  console.log(chunk.line)
}

// Non-streaming commands keep returning regular promise results.
const result = await client('project deploy')({
  args: { id: 'p1' },
})
//    ^? project deploy output

Maintainer Note: Stream Detection

Streaming typegen detection is based on actual async *run handlers. A wrapper like run() { return stream() } is not emitted as stream: true.

Changes

• Detect async generator command handlers and emit stream: true metadata in typegen.
• Type streaming client calls as async iterables while preserving existing non-streaming call types.
• Add application/x-ndjson handling for RPC stream responses.
• Preserve returned CTA/error records from async generator RPC streams.
• Cancel the response body when consumers stop reading a stream early.

Tests

• Adds client tests for NDJSON chunks, stream errors, malformed streams, missing completion records, and cancellation.
• Adds type tests for streaming command return inference.
• Adds typegen snapshot coverage for generated stream metadata.
• Adds RPC route coverage for async generator streams preserving returned CTA/error records.