Various Wizard badness

Various bad things the Wizard is doing that I've noticed. Some of these are seemingly regressions (some critical) and some aren't (?) and need to be verified. It would make sense to split this into further issues as this is getting verified, partially addressed and/or discussed, but I need padding for #9000.

• Autodetected paths are shoved into the output file unconditionally. This was introduced with the (poor) original implementation of the wizard handling of autodetected paths in 2015. 0.51.0 has multiple autodetected paths so this got worse. The intention is to add them into the existing installation list without adding them to the final file and only append the user-selected directory.
• Settings writing attempts to move the installation path to the end (the beginning, technically, the comment is misleading) of the data directories list in the case the same path was already present in the data directory list, but that doesn't do anything.
• In 0.51.0 the path (paths, currently) might not be written at all due to mysterious circumstances I haven't yet deciphered. It's not that it's impossible to do this from the Wizard but something changed that the current approach broke. In 0.50.0 this seemingly doesn't happen.
• It's possible for a directory to be duplicated if it's added again by the Wizard and I don't really understand why. This happens in 0.50 and is yet to be verified in .51; previous bullet point might make this problematic.
• It throws away existing comments. While the INI importer behavior is unrelated and debatable, I would expect the Wizard itself to merely adjust the encoding and the used data directories.
• When the INI is autodetected, it is attempted to be found in the data directory itself first, which is silly. Fixed Broken by !5196 (merged), apparently it's indeed by design because retail disc installation shoves the INI into the directory root.