ci: update Scylla test version from 2025.2 to 2026.1

[mykaul]

Summary

• Update SCYLLA_VERSION in integration test CI from release:2025.2 to release:2026.1
• Tests gated by @skip_scylla_version_lt(2026.1.0) (e.g. client_routes tests) will now execute in CI
• No code changes — CI-only

Context

The integration test suite was pinned to release:2025.2. Updating to release:2026.1 ensures CI covers the latest ScyllaDB features and catches regressions earlier.

Known issues on 2026.1: Two tests in test_client_routes.py (TestSslThroughNlb and TestFullNodeReplacementThroughNlb) have pre-existing failures that will be addressed in separate PRs:

• TestFullNodeReplacementThroughNlb: bootstrap crash due to missing rackdc properties (fix ready on fix/nlb-test-bootstrap-rackdc)
• TestSslThroughNlb: SSL connection timeout after cluster reconfiguration (under investigation)

[mykaul]

CI is complaining on The nested job 'upload_pypi' is requesting 'id-token: write', but is only allowed 'id-token: none'. - I assume some CI maintenance taking place (CC @dkropachev , @roy)

[mykaul]

AI analysis of the CI failure (did not yet review!):
Failure: All 5 asyncio test jobs fail with exactly 1 test failure each:
tests/integration/standard/test_client_routes.py::TestSslThroughNlb::test_ssl_without_hostname_verification_through_nlb
Root cause: This is an asyncio reactor SSL issue, NOT caused by the version bump itself. The version bump simply enables the client_routes tests (they were skipped on release:2025.2 since that version lacks the system.client_routes table).
Details:

• All libev jobs (6) and asyncore (1) pass this test
• All asyncio jobs (5) fail — OperationTimedOut after 5 seconds waiting for a CQL SUPPORTED response
• The asyncio reactor uses loop.sock_recv()/loop.sock_sendall() which are designed for plain sockets, not SSL sockets. When an ssl.SSLSocket in non-blocking mode encounters TLS protocol messages (e.g., TLS 1.3 NewSessionTicket), these methods don't handle the SSL renegotiation correctly
• The handle_read coroutine catches SSLWantReadError/SSLWantWriteError but the underlying loop.sock_recv() may propagate them differently than expected, or TLS 1.3 behavior in Scylla 2026.1 triggers a deadlock where the socket appears readable but recv() blocks in the SSL layer
  Recommendation: This is a known limitation of the asyncio reactor's low-level SSL approach. The proper fix would be to use asyncio.open_connection() with ssl=context or loop.create_connection() with proper SSL transport, rather than wrapping the raw socket. As a short-term workaround, the test could be marked with @unittest.skipIf for the asyncio reactor, or the connection timeout in routes_visible() could be increased.