In the paper Oracle Separation of BQP and PH, Raz and Tal exhibit an algorithm that is in complexity class BQP but not in PH.
Question: Does their proof invalidate the Extended Church-Turing Hypothesis? (Note: the Extended Church-Turing Hypothesis is not the same as the Church-Turing Hypothesis)
This is not what the paper is showing. The paper is showing a problem that can be solved in a $BQP^O$ but not in $PH^O$. This doesn't imply that BQP is different than PH. (For example, IP=PSPACE but there is an oracle such that $IP^O \neq PSPACE^O$.)
Before this paper, there was already a known oracle separation between BQP and BPP (e.g. https://cs.stackexchange.com/questions/13528/). The paper improved it to BQP and PH. While this is a breakthrough, I don't think this difference matters much for the extended Church-Turing thesis. We already knew that if we allow oracles, BQP is strictly more powerful than BPP.
The extended Church-Turing thesis is about computation without oracles. To invalidate the extended Church-Turing thesis this way, we would have to show BQP $\neq$ BPP in the normal, unrelativized world.