Support performance trace (#32973)

1. Add a OpenTelemetry-like shim-layer to collect traces
2. Add a simple builtin trace collector and exporter, end users could
download the diagnosis report to get the traces.

This PR's design is quite lightweight, no hard-dependency, and it is
easy to improve or remove. We can try it on gitea.com first to see
whether it works well, and fine tune the details.

---------

Co-authored-by: silverwind <me@silverwind.io>

modules/gtprof/trace_builtin.go

@@ -0,0 +1,96 @@
+// Copyright 2025 The Gitea Authors. All rights reserved.
+// SPDX-License-Identifier: MIT
+
+package gtprof
+
+import (
+	"context"
+	"fmt"
+	"strings"
+	"sync/atomic"
+	"time"
+
+	"code.gitea.io/gitea/modules/tailmsg"
+)
+
+type traceBuiltinStarter struct{}
+
+type traceBuiltinSpan struct {
+	ts *TraceSpan
+
+	internalSpanIdx int
+}
+
+func (t *traceBuiltinSpan) addEvent(name string, cfg *EventConfig) {
+	// No-op because builtin tracer doesn't need it.
+	// In the future we might use it to mark the time point between backend logic and network response.
+}
+
+func (t *traceBuiltinSpan) recordError(err error, cfg *EventConfig) {
+	// No-op because builtin tracer doesn't need it.
+	// Actually Gitea doesn't handle err this way in most cases
+}
+
+func (t *traceBuiltinSpan) toString(out *strings.Builder, indent int) {
+	t.ts.mu.RLock()
+	defer t.ts.mu.RUnlock()
+
+	out.WriteString(strings.Repeat(" ", indent))
+	out.WriteString(t.ts.name)
+	if t.ts.endTime.IsZero() {
+		out.WriteString(" duration: (not ended)")
+	} else {
+		out.WriteString(fmt.Sprintf(" duration=%.4fs", t.ts.endTime.Sub(t.ts.startTime).Seconds()))
+	}
+	for _, a := range t.ts.attributes {
+		out.WriteString(" ")
+		out.WriteString(a.Key)
+		out.WriteString("=")
+		value := a.Value.AsString()
+		if strings.ContainsAny(value, " \t\r\n") {
+			quoted := false
+			for _, c := range "\"'`" {
+				if quoted = !strings.Contains(value, string(c)); quoted {
+					value = string(c) + value + string(c)
+					break
+				}
+			}
+			if !quoted {
+				value = fmt.Sprintf("%q", value)
+			}
+		}
+		out.WriteString(value)
+	}
+	out.WriteString("\n")
+	for _, c := range t.ts.children {
+		span := c.internalSpans[t.internalSpanIdx].(*traceBuiltinSpan)
+		span.toString(out, indent+2)
+	}
+}
+
+func (t *traceBuiltinSpan) end() {
+	if t.ts.parent == nil {
+		// TODO: debug purpose only
+		// TODO: it should distinguish between http response network lag and actual processing time
+		threshold := time.Duration(traceBuiltinThreshold.Load())
+		if threshold != 0 && t.ts.endTime.Sub(t.ts.startTime) > threshold {
+			sb := &strings.Builder{}
+			t.toString(sb, 0)
+			tailmsg.GetManager().GetTraceRecorder().Record(sb.String())
+		}
+	}
+}
+
+func (t *traceBuiltinStarter) start(ctx context.Context, traceSpan *TraceSpan, internalSpanIdx int) (context.Context, traceSpanInternal) {
+	return ctx, &traceBuiltinSpan{ts: traceSpan, internalSpanIdx: internalSpanIdx}
+}
+
+func init() {
+	globalTraceStarters = append(globalTraceStarters, &traceBuiltinStarter{})
+}
+
+var traceBuiltinThreshold atomic.Int64
+
+func EnableBuiltinTracer(threshold time.Duration) {
+	traceBuiltinThreshold.Store(int64(threshold))
+}