fix(telegram): preserve can_edit after transient network errors in progress edits (#27828)

When edit_message_text fails with a transient error (httpx.ConnectError,
NetworkError, server disconnected, timeouts), the progress-message sender
must not permanently set can_edit = False — that would convert a single
Telegram network hiccup into separate per-tool bubbles for the rest of the run.

Changes:
- gateway/platforms/telegram.py: edit_message now returns retryable=True for
  transient network errors (ConnectError, NetworkError, timeouts, server
  disconnects, temporarily unavailable). Permanent failures (flood control,
  message-not-found, permissions) remain retryable=False.
- gateway/run.py: send_progress_messages checks result.retryable before
  setting can_edit = False. Transient failures skip the fallback-send and
  continue — the next edit cycle catches up with the accumulated lines.
  Permanent failures (flood, message-not-found, etc.) still disable editing.

Tests: 22 new tests in test_telegram_progress_edit_transient.py covering
transient vs permanent error classification, SendResult.retryable semantics,
and the can_edit decision logic.

Fixes #27828

gateway/run.py

@@ -15462,6 +15462,17 @@ class GatewayRunner:
                         )
                         if not result.success:
                             _err = (getattr(result, "error", "") or "").lower()
+                            # Transient network errors (ConnectError, timeouts)
+                            # must not permanently disable progress-message
+                            # editing — the next cycle can catch up.  Only
+                            # permanent failures (flood control, message not
+                            # found, permissions) should set can_edit = False.
+                            if getattr(result, "retryable", False):
+                                logger.debug(
+                                    "[%s] Transient edit failure — keeping can_edit=True",
+                                    adapter.name,
+                                )
+                                continue
                             if "flood" in _err or "retry after" in _err:
                                 # Flood control hit — backoff but keep editing.
                                 # Only disable edits for non-recoverable errors.