Of course it can, otherwise the chip would use an IO pin mux and time multiplex them with a single I2C peripheral. However some coding investment is needed to reach this level in bare metal. Interrupt based statemachine digesting a queue of i2c transactions.
As for 128 slave addressing per bus, you may be limited by bus total capacitancd and max bit rate, if trying to make an array of sensors for exemple. Each slave needs a unique address, and usually with limited selections.
