highway: unexpected emu128 failures

After a recent highway upgrade https://github.com/libjxl/libjxl/pull/1452, we are observing some unexpected failures in tests with the highway target emu128 on 32 bit architectures, for example

[----------] 1 test from HwyMulTestGroup/HwyMulTest
[ RUN      ] HwyMulTestGroup/HwyMulTest.TestAllMulHigh/Emu128
i16x8 expect [0+ ->]:
  0x3FFF,0x0FFF,0x03FF,0x00FF,0x003F,0x000F,0x0003,
i16x8 actual [0+ ->]:
  0xBFFF,0x0FFF,0xE400,0x00FF,0xF840,0x000F,0xFE04,
Abort at ./third_party/highway/hwy/tests/mul_test.cc:131: Emu128, i16x8 lane 0 mismatch: expected '0x3FFF', got '0xBFFF'.

see here: https://github.com/libjxl/libjxl/runs/6916936967?check_suite_focus=true Also see the discussion on https://github.com/libjxl/libjxl/pull/1500.

Two questions:

1. Shouldn’t the tests still pass when setting the tolerance to 0, as done here: https://github.com/libjxl/libjxl/runs/6917084785?check_suite_focus=true?
2. Is the HwyMulTestGroup/HwyMulTest.TestAllMulHigh/Emu128 above something that needs to be fixed upstream?