Like Pesikar points out in the comments, the audio is not really loud in the first place, but compressed differently.
That said, suppose one wanted to design a feature built into the TV that allows the viewer to limit the audio output to some specified value (as measured in dB(A) or something else across the frequency span used by the audio at that specific time). What would the challenges involved be? What would be the potential failure modes?
First, you can limit audio loudness basically in one of two ways. You can either average over a period of time and adjust the audio such that the average over that time period is at a certain value. Pick a short enough interval and this effectively limits the peaks; pick a long enough interval and it normalizes an entire movie while retaining most of the difference between valleys and peaks in the audio level. Or you can clip peaks to some predetermined loudness while doing nothing about anything else. How would either of those work out in practice?
Clipping peaks to a predetermined level is relatively easy, but it leads to distorted audio. Any time you clip anything, by definition you lose definition. So any sudden, loud, desired sounds would become distorted. Alternatively, to guard against this, producers would have to use a very low level of audio output for anything normal, forcing you to turn up the volume anyway as well as (in many cases) losing audio fidelity and still leading to a risk of clipping. This is clearly not a desirable approach.
Averaging and normalizing over a period of time would allow the TV to dynamically adjust the output volume to a predetermined level. For that, you have to pick a period of time to average and normalize over. Let's say at first you pick 50 ms, because it's a small, nice and fairly round number. So you prefetch 50 ms of audio, analyze it to determine the loudness level, and then adjust the output audio amplification before playing it back. For better and worse, 50 ms doesn't let you adjust to any longer trend, which means that this system is fairly easy to trick into using a volume other than what the viewer intended (just set the audio level to zero 1 out of 5 ms, and the peaks can rise 20% without the average being affected). If you take a longer time period, you can sort of protect against this, at the expense of a noticable audio delay (which admittedly can be corrected for by having the TV buffer video output before starting playback and basically showing a time-delayed version of the input, but that too has other problems, such as accurate timing which some people want). But then there is a scene that is supposed to be quiet: two characters whispering to each other, a nature scene, or whatever. That too will get averaged and normalized to the set value, increasing the audio output level. Suddenly, based on the audio, you can barely tell the difference between characters whispering to each other and yelling at each other, and with low audio levels, ambient sounds (as well as random noise) will get amplified to however loud you want conversations to be. Clearly not desirable! While this can be resolved by making the TV not increase but only decrease the audio level in the normalization process, doing so means it's not really outputting a more-or-less-fixed audio level anyway, so it fails your basic design criteria. Back to the drawing board.
So either way you choose to implement it, it basically fails in common scenarios. In one case it leads to poor audio quality, in the other it becomes difficult to tell how loud a sound in the movie is and increases the amount of audio noise.
How many would pay extra for that?