Fix swap of relfrozenxid, relfrozenxid and relallvisible

[za-arthur]

This PR is follow up of #157.

• I fixed the typo: Fix relfrozenxid, relminmxid and relallvisible on table data files swap #157 (comment)
• I removed PG_VERSION_NUM >= 90300 check since we don't support such old versions anyway
• I added check relform1->relkind != RELKIND_INDEX since relfrozenxid and relminmxid is non-zero only for tables

I think we still need to check the current state of swap_relation_files (https://github.com/postgres/postgres/blob/ae69c4fcf1337af399a788ab8d96af540bd1cd8e/src/backend/commands/cluster.c?plain=1#L1055). It has a few differences with the code of repack.c. We don't need to copy all the code, since not all features are supported by pg_repack. For example we don't repack system tables, mapped tables. And maybe we don't need to support swap_toast_by_content?

I'll try to come up with a PR to try to sync the code.

[MasahikoSawada]

I agree with overall of these changes, but I have some minor comments.

[MasahikoSawada]

I think we should use a MultiXactId variable instead.

[za-arthur]

I used TransactionId mainly because relminmxid has that type, regardless of the comment in pg_class.h, stating that it is a MultiXactId.
I wonder why pg_class.h doesn't use MultiXactId for relminmxid.

[MasahikoSawada]

I wonder why pg_class.h doesn't use MultiXactId for relminmxid.

I think MultiXactId doesn't have the corresponding SQL data type unlike TransactionId. Data types that can be used in system catalog are somewhat limited. It makes sense to use MultiXactId internally since we use it in C functions.

[za-arthur]

Thank you @MasahikoSawada for the review! I pushed changes.

[MasahikoSawada]

Thank you for updating the changes!

The changes look good to me. I'll merge it, barring any objections. Or if you can merge it by yourself, please go ahead.

[za-arthur]

The changes look good to me. I'll merge it, barring any objections. Or if you can merge it by yourself, please go ahead.

Thank you. I'm merging it.