Yeah, people weren't finding and reading the FAQ in the sidebar. The reason it isn't all that detailed is because it needs to be a general answer to "how do blind people use computers/reddit?" The answers to a lot of the questions your asking tend to be "it depends".
> Are things that don't work a minor inconvenience
In the case of online chat, yes. The only thing to be aware of is not to send images. Other than that, if the person you're chatting with missed something, he can ask. The only time it was ever a problem for me is when someone bolded a word to indicate sarcastic emphasis, and I had my screen-reader set not to notify me of bold words. It was obvious I'd missed the bolded word from the next message I sent, and the person I was chatting with clarified her feelings, and we moved on.
In the case of developing websites and programs, things that don't work can be a huge problem, preventing screen-readers from clicking buttons, activating controls, and so-on. However, each platform and screen-reader are so different that the best thing a developer can do is follow the applicable accessibility guidelines, and do user testing to make sure it all works on the platforms that must be supported. These days, the guidelines are good enough that if they're followed correctly, 9 times out of 10 things will work just fine.
I'm now going to go into detail about how screen-readers work, to demonstrate why none of this is in the FAQ. If you don't want massive details, stop reading now! :-)
> any special considerations when having an online chat
At the moment, the only software that can reliably read Emoji is the VoiceOver screen-reader on IPhones and mac computers. There are add-ons for the Windows screen-reader
$1 to enable Emoji support, but just like a browser add-on, you can't depend on everyone having downloaded and installed that plugin. None of the other 3 major screen-readers on Windows support Emoji. So I couldn't write something like "blind people can't read Emoji" in an FAQ, because that's not always true. Also, I haven't used an Android phone in 3 years, so it's possible Google added Emoji support to the TalkBack screen-reader and I don't know about it. Providing no information in an FAQ is better than providing wrong information, IMHO.
As for Emoticons, that's even more complicated! When a screen-reader wants to speak text, it uses a separate 3rd party module, called a synthesizer. Most screen-readers (accept for the one on IPhones) support a large number of different synthesizers, and the user can choose whatever she likes best. Each synthesizer will usually provide multiple voices (male, female, etc) and multiple languages that it can speak in. Each synthesizer has different rules to decide how to handle punctuation, how words should be pronounced, and what to do about emoticons. I know of a forum that actually had to ban people from debating about what synthesizer is "best", because people feel so strongly about the issue that the discussion will sometimes devolve into flaming and name-calling. The synthesizer is the voice that reads everything on the computer to a screen-reader user, so of course they're going to feel extremely strongly about how it should sound and what it should do. But it gets even more complicated! For various reasons, screen-readers themselves will usually modify the text, before they send it to the synthesizer. So that means that the exact same synthesizer, when used in two different screen-reading programs, will sometimes pronounce things differently! So depending on how the user has things set up, :-) might be read as "colon dash right paren", "smiley", nothing at all, "colon dash right parenthesis", or a hundred other ways. I remember the first time I used Linux, and was surprised to hear that the default screen-reader with the default settings at the time always read the exclamation mark as "bang".
> Does punctuation work well?
The screen-reader user gets to decide exactly how much punctuation she wants read, and when. For example, I have NVDA set to not read any punctuation when I'm in read all mode, but when I'm just cursoring up and down, I have it set to read "most punctuation". What punctuation marks are considered part of "most punctuation" is user configurable on a symbol by symbol basis, with a list of 130 symbols to configure. Most synthesizers will also let the user change a setting called "intonation", that controls how punctuation affects the synthesizer, even when it's not reading the symbols directly. For example, a high intonation will usually pause longer at the ends of sentences, go up higher at a questionmark, and so on.
> Does formatting work?
Depends. In software like Google Chrome, Firefox, and Microsoft Word, formatting information is given to the screen-reader. The user then gets to turn on and off announcement of different types of formatting (bold text, italics, font name, colour, block quote, links, tables, etc). But not all software supports this. And all screen-readers handle it differently. If you're making a website, just follow the W3C and other guidelines. If you're chatting with a blind user, she may or may not notice if you wrote a particular word in bold, depending on how she's got things configured.
> Do common acronyms work?
It depends on the screen-reader and synthesizer used. However, it doesn't really matter. Hearing "lol" pronounced to rhyme with bowl, or having the letters "L O L" spelled out, or having LOL pronounced as "laughing out loud" is a matter of user preference. Assuming a user knows what an acronym means, she'll be used to how her particular configuration pronounces that acronym.
> How does a screenreader handle misspelled words?
Depends on the synthesizer in use. Synthesizers use phonetic rules to decide how to pronounce words, rather than having a pre-defined pronunciation for every English word. Each synthesizer has different rules, so each one will pronounce misspellings differently. However, all screen-readers will let the user review words character by character if she's confused, so it's usually possible to figure out what terrible spellers are trying to write.