TTL indexes configured with a NaN for the expireAfterSeconds options are handled differently in 5.0+ binaries by the TTL Monitor for the purposes of calculating an expiration date for the indexed documents. To avoid a mix of 5.0 and 4.4 behaviors for these misconfigured TTL indexes, we should detect these indexes during the upgrade and downgrade steps and fail accordingly.

With this fix, MongoDB 5.0.11 will:

• Log an error and crash on start-up if it's running in FCV 4.4 and detects a bad index. In addition to clusters upgrading directly from version 4.4, this includes clusters running featureCompatibilityVersion (FCV) 4.4 in versions 5.0.0 - 5.0.10.
• Fail to create a new misconfigured TTL index (either via user operation or initial sync) when running FCV 4.4.
• Start properly and allow the creation of new misconfigured TTL indexes when running FCV 5.0, and treat this configuration as expireAfterSeconds: 0, preserving behavior from versions 5.0.0-5.0.10.
• Fail to downgrade to FCV 4.4 if a misconfigured TTL index exists.

To avoid any of these problematic behaviors, do not use expireAfterSeconds: NaN.