Miniature-scale "flame" lights do exist, but the limitation is not the size or power of the bulb; it's the power source and wiring that's the problem. Here's a visual that demonstrates the real hurdle to using these with movable 25mm scale miniatures:
Granted that's a full-featured campfire flame simulation kit with features you may be willing to do without. The circuit board handles the "flame flicker" as well the conversion of the suggested 9V power source to a voltage that an LED can accept (typically ~1.7V).
A smaller solution is possible in theory, using just a coin- or button-cell battery and skipping the circuit board. (For a bit more money, you can even get LEDs with built-in flickering effects!) The battery could be mounted under the base of a miniature. The leads would require disguising or drilling a channel through the miniature itself (the latter would be painstaking, or prohibitive on minis with small-diametre arms, but impressive). I'm not aware of any existing kits or props that supply such an arrangement, so it would be an entirely DIY kitbashing project.
In the end though, this would be almost entirely aesthetic. A normal LED will cast light far beyond the scale radius of any in-game light source, and you don't have room for circuitry to moderate its brightness to adjust that downward. Even if you did, the falloff for real light is analogue and wouldn't help to indicate the hard lines between lighting levels that most RPGs use to abstract lighting.
Their characters are fine.
Optimization in 5e has very little to do with individual characters, and everything to do with the party as a whole.
Unless the players are highly organized, builds with high stats and dump stats would actually be a liability. I'll explain this by comparing it to Pathfinder:
In Pathfinder, each character takes on a specific role in which their competency, assuming the character is optimized, grows exponentially. For example, a wizard may invest a level or two into multiclassing rogue, and it's better than nothing, but they won't hold a candle to the "real thing." At some point DCs for skill checks and saving throws get so high that only a specialized character has a real chance.
In D&D 5e, a character specializes in a number of things, but this generally only means that they are a step or two above the rest. For example, say a wizard takes the criminal/spy background; for 99% of your adventuring career, that wizard can be relied on to handle duties that are traditional to rogues, though they will likely approach said duties in a different way. No multiclassing required.
This is because of the bounded accuracy thing. You always have a chance. Having a decent wisdom score on a barbarian isn't a waste. Hello, mind control!
Because of this, it's less important in 5e that you specialize. Instead you optimize by having your party, as a whole, make sure that all bases are covered. And redundancy is more likely to save your bacon than a slightly higher bonus to your already-high rolls.
5e adventure design assumes players to be sub-optimal
To some extent, whether individually or as a cohesive unit. So you don't really need to ease up on anything. In early stages you might want to give them some light warnings when they're about to enter a deadly encounter. A wake-up call so they realize, "Oh we should play it smart" is more than enough for most parties.
As far as what I mean by "balance": it seems that at levels 1-2 it's
hard to keep everyone conscious the whole time since they have so few
hit points; better to anticipate a KO or two and bring your healing.
This is not a consequence of the characters being built for breadth. It's a simple matter of how fast and deadly combat is in 5e. Perfectly normal for any party.
Some DMs will start tuning down encounters once players start dropping to zero health. Don't do this! They survived, they learned, they will be excited, they will be tuning their party up.
That's when you start targeting players' weaknesses. At which point, it's a good thing they went for breadth.
Total party wipeouts are not the end in Curse of Strahd.
Check out the Adventure's League Dungeon Master's Guide v4 that accompanies Curse of Strahd. Look under Jeny Greenteeth's spellcasting services, and the block on "Death in Ravenloft." This is a canon method for continuing the campaign in the event of everybody dying horrible deaths.
So by all means throw that coven Night Hags at a team of level 5s! They might surprise you and curbstomp the things. Or they might all die and come back with some Dark Gifts.
Challenge Rating is accurate but not precise.
You can read about it in the DMG. CR is based on Proficiency bonus, HP, AC, attack bonus, damage per round, and DC on their abilities. It is not adjusted for the special abilities they have, and how those can play with different parties.
For example, I've run Death House a few times. The Shambling Mound is CR 5. It's supposed to be tough, but parties have had little trouble with it. On the other hand, each party was seriously thinking they were going to die when I unleashed that pack of five CR 1/2 shadows on them.
The CR does not account for how a devious DM may capitalize on special abilities, or how a given party may completely shut an enemy down.
Best Answer
Whole list of maximum of creatures in single encounter
This doesn't include named NPCs from appendix D or encounters from Death House
Death House