"Metagaming," as generally defined as using out of PC knowledge to direct PC actions (scare quotes because this is a specific definition inside the broader actual definition of metagaming), really can only exist in a situation that has a fixed fictional framing where parts of that framing are meant to be hidden from the players and discovered in play. The focus on the players here rather than the PCs is important, here. The reason for this is because if the players actually do not know these facts, then they cannot engage in "metagaming" because their reference frame is the same as their PCs. The issue occurs when the hidden structure breaks down and you have players than know the supposed hidden facts but these facts haven't been revealed to the PCs in the fiction, yet. Every argument I've seen on this relies on this set of hidden facts. The canonical example is the troll vs new players and then the troll vs experienced players. Against new players, the player knowledge aligns with the supposed PC knowledge in that neither have information on trolls, so the encounter is difficult and challenging. Against experienced players, the troll is trivialize if they use their knowledge to attack the troll's weaknesses, but this raises complaints of metagaming because it's not established that their PCs know this. Much argument has ensued.
So, "metagaming" exists when the player/PC hidden fact knowledge diverges, for any reason. Here's the controversial statement: this is mostly going to be the GM's fault, except in cases of outright cheating where a player has knowledge but conceals it from the other players for personal gain. So, outside of bad faith play (lying by omission), "metagaming" is usually a GM caused issue. It's caused by the GM establishing a fact pattern that is known by the players but expected to be not known by the PCs. You don't have to do this. You could, with a bit of effort, establish fact patterns that are unknown to both players and PCs or, alternatively, you can establish fact patterns that aren't dependent on player's knowing them. To turn back to the troll, you could reskin the troll or change it's abilities to be a surprise to both players and PCs as an example of the first, or you could just not expect the troll to be a serious single challenge to experienced players and establish that PCs do know about trolls in the latter. If you're canny, you can do the last by putting the troll in a place where fire is dangerous or difficult to use, such as a explosive gas filled chamber or underwater. This establishes a fact pattern where the players knowing about trolls is irrelevant to the anticipated challenge of the situation.
And, if you really want to drive this home, play a game where metagaming cannot exist because there's not a hidden established fact pattern. PbtA games are good for this, in that the only established fact pattern that matters is the one established in play. It's hard to metagame if there are no hidden facts for which the player/PC diverge in knowledge.
This leaves the split party table talk example. This is a situation where the divergence in knowledge is created at the table, in play. Here's a place where you can get out of your own way pretty easily. The example is given where two PCs have moved ahead and are captured/killed outside of the rest of the party's knowledge. The "metagaming" occurs if the rest of the players act on this as if they know what happened. What's the actual issue, here, though? Is it that the party will mount an operation with foreknowledge of the foe and thus trivialize the encounter? This is the same as above -- change something and it's not an actual problem. The monsters know about the party as much as the party knows about the monsters, so, while their prepping, the monster fast reaction force descends on them before their ready. Or they move, or they leave a trap. There are hordes (heh) of ways to frustrate this kind of play by just not rigidly sticking to your fiction. I get the desire to have a fixed fictional world the players engage fairly, but it's impossible to do so in the given situation because the knowledge divergence has occurred, so you can either demand that players act against their play goals and risk PCs in ways they don't want to or you can change with the situation a bit. It doesn't have to be much. You can also arrange to have the PCs find out what happened in myriad ways. This kind of problem occurs because GMs have decided to codify into fixed forms areas that the game has left open and have painted themselves into corners. The game rules do not define an exact reality, they provide a general outcome and leave open large areas for interpretation and improv.
Metagaming is caused by an insistence on a fixed, immutable fact pattern that is intended to be hidden from the PCs but is not hidden from the players. You can correct for this by changing the fixed and immutable part, the hidden from PCs part, or the not hidden from players part. Or you can complain that people tend to actually act on what they know instead of pretending otherwise and be upset when it happens at your table because you've set up the conditions for it -- and blame the players for it. I don't let my players metagame, not because I insist that they ignore things they know, but because I, as GM, don't set up the conditions for it to exist. If I do, I blame myself, not the players.