A game that supports a larger range of player numbers will appeal to more people, and will more easily get published.
Your game should support 2-5 players, if possible. Most games get very slow with five players. If it's a heavier game, just stick to 2-4.
Also consider manufacturing costs. If each player has lots of personal components, a game might top out at four players, instead of five. Four is the minimum for a normal game.
Your game is going to be very hard to playtest if it doesn't work with two players. 2-4 should be your default range, but see if you can add 5.
You should support 4 players, unless you have a good reason not to. Getting a game published is harder if it doesn't. I've done it. The most important — and normal — range is 2-4. It's also fine to do just 3-4, if your game has trading or voting, for instance, and doesn't work at 2.
6-player games are so slow that they shouldn't be made, without designing the game around them. The game turns should be super-fast, or highly-interactive (such as in a trading game.) Alternatively, they could be team-based, or have simultaneous turns (or no "turns" at all.)
Scaling isn't an exact science. The game will be somewhat distorted at different player counts, regardless of what you do.
For standard 2-4 player games, I try to optimise for 3 players, so that 2 and 4-player are both close to the mark.
If your game is very adversarial, just make it a two-player game. Don't sacrifice the gameplay, to add players. Two-player-only games are harder to sell, but do what's right for your game regardless. I wouldn't design a game that requires any other specific number of players, however.
I was having significant trouble getting Radlands published, and some publishers were telling me that they weren't interested in a two-player-only game. I added modes for 1 to 4 players, even though it was really a very adversarial two-player game.
Roxley liked my game, and agreed to publish it. I brought up the 1-4 player modes I'd added. They said that sometimes it's best for a game to just support the player counts it's intended for. They hadn't done this with their game Santorini, which was really just a two-player game.
I replied "Oh, yeah. I played that game, with three players. Then, I went and rated it 5/10 on BoardGameGeek!"
A small number of players will want to play your game solo.
Unless your game is designed for it, I don't think a solo variant is worth your time.
I had some ideas about 1-player rules for Radlands, but never did anything with them. The fans of Radlands came up with solo rules that are probably much better than anything I could throw together.
Regardless of the player count, design for two-player first, so you can playtest. Then, add more players later on.