[GR-62119] Remove AccessAdvisor assertions with data races.

[graalvmbot]

In CI, we hit cases where these assertions tripped. Each call to shouldIgnoreClass is definitely preceded by a call to shouldIgnore that returned false, so this was odd.

AFAICT, the only non-final value used in the body of shouldIgnore is the "is live" check. My suspicion is that we receive a VMDeath event on a separate thread, set "is live" to false, and then the second call to shouldIgnore returns true.

We don't actually do any synchronization when the VM phase updates, which is likely a more fundamental problem with the agent. Rather than try to tackle that here, I propose to remove the assertions (I have manually checked that calls are preceded by shouldIgnore checks) to fix the transient errors.