First, although it is common in some textbooks, I don't think it is a good thing to necessarily relate the equiprobability postulate to ergodicity.
Second, what this postulate enables is to estimate the probability distribution for the macrovariable you want to look at. You can of course look at the most probable value for this macrostate and this will correspond to a "thermodynamic interpretation" of what you can expect to observe.
However, in statistical mechanics, what you can expect is not the most probable value but rather the average value and they need not be the same.
Moreover, you may want more than simply the average value; you can also want to predict what is the free energy difference between one value of the macrostate and another in some kind of transformation in your sytem and for this you are in trouble if you don't try to get the probabilities right.
Suppose you have a box of volume $V$ filled with a mol of an ideal gas with internal energy $E$. This defines the macrostate of your system, or intuitively, how your system looks in a macroscopic scale. However, we still don't know how it looks in a microscopic scale, i.e., we don't know how the $\sim 10^{23}$ particles over there are behaving individually. There are many different possibilities, which are the microstates of the system. For example, at time $t = t_0$ they could have positions $x_i$ and velocities $v_i$, where the index runs over all the particles. This is one particular microstate. However, the macrostate would be the same if particle $i=1$ had position $x_2$ and velocity $v_2$ while particle $i=2$ had $x_2$ and $v_2$ (I'm assuming things are classical and indistinguishable for simplicity). So which is the correct microstate?
From a macroscopic point of view, we don't know. All we can do is attribute what is the probability of the system being in each possible microstate. The principle you stated implies that both microstates I exemplified are equally likely to be the actual microstate. We don't know which is the right microstate, and all the possible ones are equally likely.
The system moving towards the largest number of microstates is then not only a change of microstates, but also a change of macrostate. If I mix my gas with another box of gas at different temperature or something, the system will reach equilibrium at the macrostate with the most possible microstates. We still won't know what is the right microstate, being able only to attribute probabilities.
Essentially, as OP pointed out in the comments, the idea is that since we do not know which microstate is the correct one, we assign equal probabilities to all of them.
Now, this does have a bit of nuance. Is it always valid to do this? In fact, it depends on the information you have about your system. Instead of an ideal gas, let us pick a generic gas. If the energy is fixed, then all available microstates should have the very same internal energy and there is no reason to prefer one of them over the other ones. We call this the microcanonical ensemble. On the other hand, suppose temperature (which is related to the expectation value of energy) is fixed. In this situation, there could be states with more internal energy than others, as long as the temperature stays the same (for the ideal gas, this won't happen because the energy is proportional to temperature, but let us consider a more general scenario). In this situation, it can be more likely for microstates with lower internal energy to occur, so we won't pick all probabilities to be the same. Instead, they are given by a Boltzmann distribution. This is known as the canonical ensemble.
The key point is that since we do not know what is the true microstate, we can only assign probabilities. We do this according to the information we have (or according to the experimental conditions, if you prefer). For fixed energy, all microstates should have the very same probability of being the true microstate, so they are, in this sense, equally likely.
Best Answer
"The microstate" is basically all the information you need to completely specifiy your system. For example when you have $N$ particles this would mean you need $N$ positions and $N$ velocities. The phase space is $\mathbb R^{2N}$ which can be incredibly large. For $N=1$ you can draw the complete trajectories in phase space but this could hardly be called statistical mechanics. To perform time evolution you evolve the particles under the equations of motion. For classical particles this just means to apply Newton's laws.
Another example is a system of $N$ coins which can each be head or tails. The phase space is $(\mathbb Z_2)^{N}$ where $\mathbb Z_2=\{0,1\}$. At each timestep you flip one coin at random. Now "time evolution" means to take one timestep because time is now discrete.
Macrostates are what you get when you forget all microscopic details. For a microcanonical ensemble you only need $N,V,E$ to describe the system. All these 3 variables are fixed. When you say something like NVE-ensemble you mean that all these variables are fixed and can be used to describe the system.
Note that when you are in equilibrium the macrostate doesn't care about time evolution.