When a network middlebox silently drops an idle TCP connection, the next request raises a ConnectionError. Previously the retry logic would reuse the same dead session, failing up to 10 times. Now we close and null out the session on ConnectionError so ensure_session() creates a fresh one on the next retry.

This differs from #854 in that it reuses the existing error_retry machinery rather than adding a parallel retry system with separate timeout constants and retry counters.

Fixes: #761

How did you test this PR?

Added a new automated test in zulip/tests/test_do_api_query.py that simulates a stale connection by having the first session raise a ConnectionError, then verifies that the session is closed and replaced with a fresh one, and that the request succeeds on retry.

• Self-reviewed the changes for clarity and maintainability
  (variable names, code reuse, readability, etc.).

Communicate decisions, questions, and potential concerns.

• Explains differences from previous plans (e.g., issue description).
• Highlights technical choices and bugs encountered.
• Calls out remaining decisions and concerns.
• Automated tests verify logic where appropriate.

Individual commits are ready for review (see commit discipline).

• Each commit is a coherent idea.
• Commit message(s) explain reasoning and motivation for changes.

Completed manual review and testing of the following:

• Visual appearance of the changes.
• Responsiveness and internationalization.
• Strings and tooltips.
• End-to-end functionality of buttons, interactions and flows.
• Corner cases, error conditions, and easily imagined bugs.