Merge Videos.

Not if the clips were encoded the same way. Two parts of one recording share a codec, a profile, a resolution, a frame rate, and an audio layout, and in that case the tool appends the compressed packets with a timestamp offset: nothing is decoded and nothing is encoded, so the output holds the same picture that went in. That is the multi-part phone case and it is the common one. When the clips disagree there is no way around a re-encode, because two different shapes cannot be one file, and then the output is a single fresh encode at a generous bitrate rather than a copy of a copy. The plan line tells you which of the two is about to happen before you press the button, so you are never guessing.

They land on one canvas. By default the canvas is the largest clip, measured by pixel count of the frame as it displays, so nothing is scaled up more than it has to be; you can switch it to the first clip if the opening shot is the one that should set the shape. Smaller clips are then letterboxed inside that canvas, keeping their aspect ratio with black either side or above and below, and the plan names exactly which clips got bars. If you would rather fill the frame, crop to fill scales each clip until the canvas is covered and loses whatever hangs over the edge. There is no third option that keeps every pixel and one shape, and any tool that implies otherwise is stretching your footage.

Yes, and the result is honest rather than pretty: one of the two orientations ends up boxed. Pick the canvas that matches the platform you are posting to, put the clip whose shape you want first and choose to match the first clip, and the others fit inside it. A portrait clip on a landscape canvas gets bars either side, which is what every feed does to it anyway; a landscape clip on a portrait canvas gets bars above and below unless you crop to fill, which will cut the sides off a wide shot. Rotation flags are applied before any of that, so a phone clip that reports itself sideways stands upright on the canvas rather than lying down in the middle of your edit.

Twenty clips, and no fixed limit on their length. The ceiling is about the browser rather than about video: every clip is probed on the way in and each one holds a decoder while it is being read, and a tab with twenty of those open is a tab close to falling over. Length is different, because the output is written straight to browser storage as it goes rather than held in memory, so joining a handful of long clips is a question of time rather than of space. On the lossless path that time is roughly what copying the files would cost. On the re-encode path it is one encode of the total length, and the progress bar tells the truth about where it is: converting, encoding, then muxing.