Mouser's website isn't working for me at the moment. Here's the
MITI-3V1-6-8 datasheet direct from Littelfuse.
As I suspected its a Form A (N.O.) reed switch and from your description, the magnet isn't activating the switch when you want it to be in power-save mode.
I wouldn't rely on a 2Meg resistor to hold a GPIO pin at a valid high or low level. The datasheets you linked specifies the input leakage current as max. 50nA, but it would be very difficult to achieve that in real life due to board surface leakage. 50nA through 2Meg is 0.1V. The ESP32 has 25% & 75% (of Vcc) input thresholds. At 3.3V, that's 0.825V & 2.475V so with a 2Meg pulldown you will run into trouble if the total leakage is over 0.4uA, and will get false triggering due to noise pickup long before that the upper limit
The ESP32 internal pullups/pulldowns are nominally 40K, so one strategy would be to poll closed switches with a 1:50 duty cycle, which would give you comparable current consumption to using 2Meg external pulldowns. As the input must not be left floating, in-between polls, it should be pulled to (or driven to) the same level as the switch connects it to. Personally I'd wire the switches as active low, as it decreases the risk of any short-circuits causing damage.
Depending on wiring length and screening you may need low-pass filtering at the input pins to achieve acceptable EMI immunity. Littlefuse do not recommend in excess of 100nF directly across a reed switch:
https://m.littelfuse.com/~/media/electronics/application_notes/reed_switches/littelfuse_magnetic_sensors_and_reed_switches_capacitive_loads_application_note.pdf.pdfand in a low duty cycle switched pullup application, you also have to be concerned about the RC time constant with the internal pullup, so the size of the filter capacitance and the required duty cycle determine the max polling rate. How much lag between switch opening and its detection can your application tolerate?
Also "A Guide to Debouncing, or, How to Debounce a Contact in Two Easy Pages", by Jack Ganssle is essential reading:
http://www.ganssle.com/debouncing.htm