1) to measure a resistor, the meter is outputting a known current at a unknown voltage, and measuring the voltage
2) meter reading 10V across 10Mohm can extrapolate as 1uA current. or if known source voltage is 100V can extrapolate Rx = 90Mohm
Multimeters use an "ohms converter" circuit which is simply a precision current source applied to the output when in voltage measurement mode. The output current is scaled by the range and the basic voltage measurement, typically 0.2000 or 2.000 volts for a 3-1/2 digit instrument, is used to make the measurement. Since the input decade divider is used to scale the current, it is unavailable as a 1 or 10 megohm load to the input, so the input resistance is infinity and does not interfere with the resistance measurement.
I am not sure about the exactly configuration for autoranging multimeters which use a singled ended decade divider, but they do something similar.
The sampling capacitor is charged to the input voltage resulting in a charge proportional to voltage, and then charge redistribution is used to measure the charge.
They do indeed charge the sample capacitor to the input voltage, but in essence it is still a sampling capacitor, which means very high DC input impedance, but short current peaks during sampling, especially when combined with a MUX and sampling different input voltages.
I was only discussing charge redistribution ADCs which are commonly used in microcontrollers and known as successful approximation converters, although this also covers delta-sigma converters which have a similar input structure. There are now some buffered delta-sigma converters which have low charge injection and work with moderate (10k) impedance sources.
In the end you have to always take all the peculiarieties of your ADC into consideration when designing a frontend.
With these charge redistribution things for example, you can put a 10Meg series resistor in front of it if you combine it with a capacitor to make a low-pass filter. The capacitor should then be several orders of magnitude bigger then the sampling capacitor, so it coompletely swamps it, but as the sampling capacitor is typically in the pF range this is easy to do.
With some early exceptions, multimeters buffer their input but charge injection from the input multiplexer still exists. The input multiplexer is required to implement the automatic zero loop for the input buffer to remove offset and common mode errors. Some early multimeters have a separate precision buffer outside of the automatic zero loop however this places considerable demand on the performance of the buffer.
The early exceptions are interesting because their discrete analog-to-digital converter is high impedance without a buffer or multiplexing so they do not suffer from any charge injection on their input which can be useful in certain applications. How do you make a multimeter with no multiplexing, no precision buffering, and still achieve better than 100 microvolt precision? They did it by using a non-inverting operational integrator, which serves as the integrator for the analog-to-digital converter, while the non-inverting input of the operational integrator is the high impedance input. The only special part required is a JFET differential pair.
Measuring resistance directly is impossible. The only way that any instrument (DMM or otherwise) can do it is to supply a voltage, measure both the voltage and the current, then do the math for you.
I have measured resistance before by measuring the Johnson noise from the resistor. I did this testing low noise DC amplifiers, and no applied voltage or current was required. Accuracy was comparable to a good handheld multimeter.