feat(tests): add streaming HTTP integration test and example components

[Aditya1404Sal]

Feature or Problem

This PR adds comprehensive testing and example components for streaming HTTP responses in wasmCloud, providing:

• Integration tests to verify streaming behavior
• Example components demonstrating real-world streaming use cases
• Documentation through working examples

The streaming capability is essential for:

• AI/LLM applications that stream generated text token-by-token
• Large file downloads with progress feedback
• Server-Sent Events (SSE) style responses
• Any scenario requiring low time-to-first-byte

Related Issues

• Builds on fix: call http handler in separate task #195 (Store lifetime management for streaming)

Release Information

Target: next release

Consumer Impact

Breaking Changes: None

Benefits:

• Validated streaming behavior through comprehensive integration tests
• Reference implementations for developers building streaming components
• Documented patterns for AI streaming and SSE-style responses

Testing

Unit Test(s)

• No new unit tests (streaming support tested via integration tests)

Acceptance or Integration

Added comprehensive integration test: tests/integration_http_streaming.rs

Test Design:

• Uses a simplified streaming component that generates 20 random data chunks at fixed 0.5-second intervals
• Validates precise timing and chunk count requirements for reliable CI/CD testing
• Independent of external API dependencies (no API key expiration failures)

Example Components Included:

1. http_streaming.wasm - Deterministic test component for CI/CD
2. http_ai_proxy.wasm - Real-world Gemini API proxy demonstrating AI streaming use cases

Test Assertions:

• Verifies exactly 20 chunks are received
• Validates total streaming duration is approximately 10 seconds (9-11s range)
• Confirms each chunk arrives at ~0.5 second intervals (0.4-0.6s range)
• Ensures response is non-empty and properly formatted
• Tests that streaming works correctly

Why this approach?
The integration test prioritizes reliability and determinism over realism. Using a controlled streaming pattern ensures tests won't fail due to:

• Expired API keys
• Network connectivity issues
• External service rate limits
• Variable upstream response times

The real-world Gemini proxy example is included to demonstrate practical streaming applications, but the CI test uses a predictable pattern.

Manual Verification

1. Built and ran both components:

   • http_streaming.wasm (test component with deterministic timing)
   • http_ai_proxy.wasm (Gemini proxy example for real-world use)

2. Deterministic Test Component:

✓ All streaming assertions passed!
  - Received exactly 20 chunks
  - Total time: 10.09s (~10s expected)
  - Average interval: 0.504s (~0.5s expected)

3. Gemini Proxy Example (Real-world streaming):

• Sent POST request with prompt: "Explain how AI works"
• Observed streamed response chunks arriving progressively
• Verified components can continue streaming after the handler returns
• Results:

✓ 29 chunks received
✓ 15,355 bytes total
✓ 9.83 seconds total duration
✓ Response includes valid SSE-formatted Gemini API data
✓ Content-Type: text/event-stream
✓ Transfer-Encoding: chunked

Observed streaming behavior (Gemini example):

• First chunk arrives at 10μs (immediate header response)
• Subsequent chunks arrive every 300-500ms (realistic streaming from Gemini API)
• Average chunk size: ~530 bytes per chunk
• Total response: 1,512 tokens streamed progressively

Both approaches confirm that the streaming implementation works correctly and components can continue streaming after the handler returns.

[thomas9911]

I dont know if LocaResources implemented Default, might be cleaner to use that instead:

                local_resources: LocalResources {
                    allowed_hosts: vec!["generativelanguage.googleapis.com".to_string()],
                    ..Default::default()
                }

[Aditya1404Sal]

Yes, LocalResources::default() is present

[Aditya1404Sal]

CC @lxfontes pretty sure the CI failure is a flake, So it's ready for a review

[Aditya1404Sal]

@if0ne it seems that #195 had the same changes present in this PR, do you reckon if I should keep the streaming examples/moreover this entire PR open or just close it ?

[if0ne]

I like the examples with streaming. Rebase on upstream/main.

[if0ne]

You can make the generated async trait and get rid of the code related to the executor

https://docs.rs/wit-bindgen/latest/wit_bindgen/macro.generate.html#options-to-generate

[Aditya1404Sal]

That's neat! didn't know that was possible 😮

[Aditya1404Sal]

@lxfontes I've refactored the fixture part here and rebased, should be good for a review!