I felt the same, but then thought of experts in their field. For example, my PhD advisor would already know all these papers. For him the prompt would actually be similar to what was shown in the video.
You are absolutely right, and a follow-up CL fixed the non-Zba code path. At the time of writing of the blog post, the generated code was however still using the 4 instructions instead of just two.
V8 is a huge project and ports almost have to start from backends of existing architectures. Over time we (I recently worked on the RISC-V port) improve the situation.
Note that it's also crucial to stay somehow similar to the x86 and ARM backends as the V8 team frequently makes changes to these and keeping up with them would be a nightmare if the RISC-V backend was too different.
"Does no one at Google know RISC-V"
This work wasn't done by Google. Also, you don't hire RISC-V experts, but V8 experts (or other similar VMs) for such a port. I, for example, had never worked with RISC-V before. As such, it can happen that some code of the RISC-V backend is written by engineers that aren't yet experts in RISC-V.
When I published Grisu (Google double-conversion), it was multiple times faster than the existing algorithms. I knew that there was still room for improvement, but I was at most expecting a factor 2 or so. Six times faster is really impressive.