feat: add aibtest package for testing utilities

[ibetitsmike]

Summary

This PR creates a proper testing package to address the tech debt identified in #73. The package consolidates scattered test helpers and provides well-structured components that reduce boilerplate and improve test readability.

New Components

MockUpstreamServer

Simulates LLM provider API responses using txtar fixtures. Features:

• Automatic streaming/non-streaming response handling
• Response mutation for multi-turn conversations
• Custom status code support
• Call counting

MockMCPServer

Provides a mock MCP server with integrated call tracking. This addresses the callAccumulator issue mentioned in #73:

• Combines server, HTTP endpoint, and call accumulator into a single struct
• GetToolCalls(toolName) returns all recorded invocations
• TotalCallCount() returns total invocations across all tools
• NewMCPManager() creates an initialized ServerProxyManager ready to use

MockRecorder

Implements aibridge.Recorder for testing with convenient accessors:

• Interceptions(), TokenUsages(), ToolUsages(), PromptUsages()
• TotalInputTokens(), TotalOutputTokens()
• VerifyAllInterceptionsEnded(t) for validation

TestBridge

The main orchestrator that combines everything:

• Creates bridge, HTTP server, recorder, and optional MCP server
• DoAnthropicRequest(t, reqBody) and DoOpenAIRequest(t, reqBody) methods
• Automatic cleanup on test completion

Example Usage

Before (from bridge_integration_test.go):

// Manual txtar parsing
arc := txtar.Parse(fixture)
files := filesMap(arc)
reqBody := files[fixtureRequest]

// Manual server setup
ctx, cancel := context.WithTimeout(t.Context(), time.Second*30)
srv := newMockServer(ctx, t, files, nil)
t.Cleanup(srv.Close)

// Manual recorder setup
recorderClient := &mockRecorderClient{}

// Manual MCP setup with separate callAccumulator
mcpProxiers, acc := setupMCPServerProxiesForTest(t, testTracer)
mcpMgr := mcp.NewServerProxyManager(mcpProxiers, testTracer)
require.NoError(t, mcpMgr.Init(ctx))

// Manual bridge setup
logger := slogtest.Make(t, &slogtest.Options{}).Leveled(slog.LevelDebug)
providers := []aibridge.Provider{aibridge.NewAnthropicProvider(cfg, nil)}
b, err := aibridge.NewRequestBridge(ctx, providers, recorderClient, mcpMgr, logger, nil, testTracer)

// Manual HTTP server setup
mockSrv := httptest.NewUnstartedServer(b)
mockSrv.Config.BaseContext = func(_ net.Listener) context.Context {
    return aibridge.AsActor(ctx, userID, nil)
}
mockSrv.Start()
t.Cleanup(mockSrv.Close)

// Make request
req := createAnthropicMessagesReq(t, mockSrv.URL, reqBody)
resp, err := client.Do(req)

// Verify tool calls via separate accumulator
invocations := acc.getCallsByTool(mockToolName)

After (with aibtest):

// 1. Create mock upstream
upstream := aibtest.NewMockUpstreamServer(t, ctx, fixture)

// 2. Create MCP server with integrated call tracking
mcpServer := aibtest.NewMockMCPServer(t, aibtest.TestTracer())

// 3. Create test bridge (combines everything)
bridge := aibtest.NewTestBridge(t, ctx, aibtest.TestBridgeOptions{
    Provider:  aibridge.NewAnthropicProvider(aibtest.AnthropicConfig(upstream.URL, aibtest.DefaultAPIKey), nil),
    MCPServer: mcpServer,
})

// 4. Make request
resp := bridge.DoAnthropicRequest(t, upstream.Files()[aibtest.FixtureRequest])

// 5. Verify - tool calls are tracked in mcpServer directly
calls := mcpServer.GetToolCalls("my_tool")
bridge.Recorder.VerifyAllInterceptionsEnded(t)

Issues Addressed from #73

• ✅ callAccumulator is now integrated into MockMCPServer
• ✅ setupMCPServerProxiesForTest return values are consolidated
• ✅ proxies return value complexity is eliminated
• ✅ Common test patterns are abstracted into TestBridge
• ✅ Package can be used externally (e.g., in coder/coder integration tests)

Testing

All existing tests continue to pass:

ok      github.com/coder/aibridge       0.186s
ok      github.com/coder/aibridge/aibtest       1.346s
ok      github.com/coder/aibridge/buildinfo     0.004s
ok      github.com/coder/aibridge/mcp   0.008s
ok      github.com/coder/aibridge/utils 0.004s

Resolves #73

[dannykopping]

Superseded by #98