feature/taildrop: replace outgoing-file progress channel with synchronous reporter

serveFilePut tracked outgoing-file progress through an unbuffered
progressUpdates channel whose close was owned by the request goroutine
while writers were spread across manifest parsing, the
progresstracking.Reader callback, singleFilePut failure paths, and the
success path. That writer-closes mismatch made the
send-on-closed-channel panic effectively unfixable in place.

Replace it with a request-scoped outgoingProgress reporter. Transfer
code reports state by method call; the reporter coalesces hot-path
updates and is flushed once via defer in serveFilePut. With no
producer channel to close, the panic is structurally impossible.

Fixes #19115
Fixes #19817

Change-Id: I8f00d982d2c79880dfc1f8104c5eed06e94b5a6c
Signed-off-by: James Tucker <james@tailscale.com>

feature/taildrop/ext.go

@@ -9,7 +9,6 @@ import (
 	"errors"
 	"fmt"
 	"io"
-	"maps"
 	"path/filepath"
 	"runtime"
 	"slices"
@@ -411,14 +410,16 @@ func (e *Extension) taildropTargetStatus(p tailcfg.NodeView, nb ipnext.NodeBacke
 	return ipnstate.TaildropTargetAvailable
 }
 
-// updateOutgoingFiles updates b.outgoingFiles to reflect the given updates and
-// sends an ipn.Notify with the full list of outgoingFiles.
-func (e *Extension) updateOutgoingFiles(updates map[string]*ipn.OutgoingFile) {
+// updateOutgoingFiles merges updates into e.outgoingFiles and emits an
+// ipn.Notify.
+func (e *Extension) updateOutgoingFiles(updates map[string]ipn.OutgoingFile) {
 	e.mu.Lock()
 	if e.outgoingFiles == nil {
 		e.outgoingFiles = make(map[string]*ipn.OutgoingFile, len(updates))
 	}
-	maps.Copy(e.outgoingFiles, updates)
+	for id, f := range updates {
+		e.outgoingFiles[id] = &f
+	}
 	outgoingFiles := make([]*ipn.OutgoingFile, 0, len(e.outgoingFiles))
 	for _, file := range e.outgoingFiles {
 		outgoingFiles = append(outgoingFiles, file)