test: introduce planner and optimizer test

[skyzh]

For database developers, query plans are important -- whether a change in optimizer will bring unexpected result? how is a query being processed currently? is every optimization rule work as expected?

To help visualize the result of optimizer, I propose to have a planner test framework in RisingLight.

Planner test will be defined with yaml files in the following format:

- sql: |
    /* TPC-H Q6 */ 
    create table lineitem(...);
    select
      sum(l_extendedprice * l_discount) as revenue
    from
      lineitem
    where
      l_shipdate >= date '1994-01-01'
      and l_shipdate < date '1905-01-01'
      and l_discount between 0.08 - 0.01 and 0.08 + 0.01
      and l_quantity < 24;
  logical_plan: |
    <the logical plan>
  physical_plan: |
    <the batch plan>

For each item in the yaml file, there will be a sequence of SQL statements, where the last statement will be used to do plan test. We will serialize logical plan the physical plan into the yaml, commit them into the git repo, and test if the current optimizer generates the same plan as test cases, so that we can easily compare the changes before and after a PR.

Our plan nodes already supports serialization with serde_erased, so serialization should be easy.

At the same time, we'd better also have a utility command to update the result of all plan test cases, like cargo run --bin apply-planner-test.

[likg227]

It's very similar to RisingWave's planner test framework, we can port it to RisingLight after RisingWave open source.

[skyzh]

So it's time to port https://github.com/singularity-data/risingwave/tree/main/src/frontend/test_runner to RisingLight 🤣

[skyzh]

fixed by #661