I expect it to work well and no damage to the processor may happen:
The I/O pins have protection diodes to Vcc. When the 5 V pull-up tries to pull an I2C line over Vcc the diodes will clamp the voltage to Vcc + diode drop. About 3.9 V in this case. This also means the HIGH on the bus won't be marginal 3.3 V but more reliable 3.9 V. And the "injection current" is mere (5 - 3.9)/4k7 ~ 200
uA. The pins of "blue pill" are specified for 5 mA. Should be OK even considering slightly worse performance of a possible clone.
Of course there must be something that uses this injected 400 uA on the 3V3 rail. Unless some very low power sleep mode is used I don't think it is an issue.
I don't want to say the I2C bus is OK. I2C + noise = problems! I just think the noise susceptibility is the only problem of the setup that won't be fixed by level shifters/change of devices. I would consider SW fix to recover from I2C errors.