Fix truncation of device addresses 0x40 and above + various enhancements

[tonyle72]

This update makes several fixes and improvements to the I2C master:

• Fix truncation of device addresses 0x40 and above, caused by dropping of the saved_device_address MSB during barrel shifting. The testbench targets only one device address below 0x40, which is why this bug hasn't been caught until now.
• Remove pulsing of serial_data_output_enable during S_READ_REG, to reduce wire contention during slave-output operation.
• Update the testbench to instantiate and target two slaves at addresses 0x11 and 0x41, to verify the address-truncation fix.
• Reclock busy to always mirror (state != S_IDLE). This allows busy to be used as a true not-ready indicator. Transaction parameters are now guaranteed to be captured when (reset_n && enable && !busy).
• Use saved_device_address and saved_read_write (rather than the live inputs) to prep for slave-data receive, preventing any input changes outside of enable from sending a read operation astray.
• Extend the low portion of SCL from a quarter to a half cycle during START.