feat: Enable creating dynamic filters for both sides; support right side bound collection

[crystalxyz]

Which issue does this PR close?

• Part of a series of PRs for Enable dynamic filter pushdown for non-inner joins in physical plans #16973.
• Built upon Add JoinType preservation helpers and dynamic_filter_side; enable dynamic filter pushdown in HashJoinExec #17518 with little refactoring

Rationale for this change

The first step to fully support dynamic filter pushdown is to allow dynamic filter creation when the build side is on the right. This feature enables right side as the build side, and also implements bound collection from the right.

The concept "build side" here means that either this side is preserved, or its bound can be safely pushed down to the other side without impacting the result. The other side is called "probe side".

What changes are included in this PR?

• A function that determines which side is the build side for all the JoinTypes.
• Support the creation of dynamic filter for more JoinTypes.
• Add support for the joins that have right build side to collect stream correctly.
• Bound collection for right build side (via stream).

Note: In hash join, we collect the left side at first and get its bound naturally. For right build side, it needs to go through a different path by calculating bounds in fetch_next_probe, which is not the best practice.

TODO: Still looking for ways to make right side bound collection more natural.

Are these changes tested?

Unit test added to test both left join and right join. Bounds are collected correctly.

TODO: verify if any query plan is impacted.

Are there any user-facing changes?

No user-facing changes because the physical plan implementation should be internal.

No breaking change to public APIs.

[jonathanc-n]

cc @adriangb

[jonathanc-n]

I may be wrong here, but I believe hash join exec creates any dynamic filters any swapping, so the right side will always be the probe side.

[crystalxyz]

I may be wrong here, but I believe hash join exec creates any dynamic filters any swapping, so the right side will always be the probe side.

@jonathanc-n Yeah thanks for pointing that out. I found that I didn't update the probe_side() function. I will try to refactor it, although the assumption (that the right side is always the probe side) is deeply embedded in the code.