Convert IOContext, MergeInfo, and FlushInfo to record classes

[uschindler]

Followup to #13204: This converts IOContext to a record class. This has several positive effects:

• equals, hashCode and toString are autogenerated by an invokedynamic, so we can't forget to handle it correctly
• it is also immutable
• all parameter checking logic can be moved to some special constructor without any parameters (this is new syntax since Java 17)

The IOContext has some crazy constructors, I left them in, but all have to delegate to the default constructor now.

[uschindler]

I needed to change some consumers to call the getter methods instead of accessing fields (which are now hidden).

[uschindler]

We should possibly convert other classes to records, too (like FlushInfo or MergeInfo if they are immutable).

[uschindler]

I had to fix the missing-doclet javadoc parser to understand records. The easy fix was to treat them as "classes".

[uschindler]

I also converted the MergeInfo and FlushInfo classes to records. Very clean now.

[rmuir]

thanks for fixing the missing-doclet to deal with records!

[uschindler]

thanks for fixing the missing-doclet to deal with records!

This was a requirement, because building Javadocs failed without that fix. It is really nice that the code throws an exception when it hits an unknown element!

[uschindler]

I will apply those changes after #13196, because if doing this before, backporting of the other PR gets harder (#13196 has changes to IOContext, too).

[jpountz]

Much nicer!

[ChrisHegarty]

LGTM

[uschindler]

Hi @ChrisHegarty,
can you have another look on the crazy randomAccess flag afetr I merged main into this. Especially the checks in the record's constructor should be checked again. I have no idea what flags are mutually exclusive. Maybe there are more combinations missing.

Luckily @jpountz opened #13211, can't wait to see this be resolved.

[ChrisHegarty]

Hi @ChrisHegarty,
can you have another look on the crazy randomAccess flag afetr I merged main into this. Especially the checks in the record's constructor should be checked again. I have no idea what flags are mutually exclusive. Maybe there are more combinations missing.

The record constructor checks, while not pretty, seem right to me. This is a nice cleanup and somewhat easier to reason about.