SpecificLossPatterns tests fail on msquic with openssl TLS backend

[nhorman]

Describe the bug

The SpecificLossPatterns tests fail on msquic with the openssl TLS backend when being run with XDP/QTIP on windows.

Specifically, the addition of ML-KEM keyshares on openssl increases the size of the client/server hello messages, causing them to cross udp datagram boundaries. This in turn increases the time to complete individual handshakes that trigger loss recovery, extending the overall test run time to be longer than the configured 60 second threshold that triggers the test watchdog timeout.

Currently this is worked around by disabling ML-KEM in tls_openssl.c via a call to SSL_set1_groups_list().

this issue should be investigated to determine the root cause of the additional recovery time, and either correct that, or understand that the additional time to recover from loss is needed for larger keyshares and adjust the test timeout accordingly.

Once that is complete, the workaround above can be removed so that ML-KEM support can be restored.

Affected OS

• Windows
• Linux
• macOS
• Other (specify below)

Additional OS information

Note that this only seems to occur when QTIP/XDP is used. Unsure why that is the case.

MsQuic version

main

Steps taken to reproduce bug

1. Build msquic with -Tls openssl
2. Run the HandshakeSpecificLossPatterns tests with -UseXDP and -UseQtip enabled.

Expected behavior

Test passes

Actual outcome

Test times out due to test watchdog timer expiring.

Additional details

No response