feat(small): Support <slt:ignore> marker in sqllogictest for non-deterministic expected parts

[2010YOUY01]

Which issue does this PR close?

Part of #17612

Rationale for this change

sqllogictests are in general easier to maintain than rust tests, however it's not able to test EXPLAIN ANALYZE results, because their results include changing part:

(in datafusion-cli) The elapsed_compute measurement changes from run to run.

> EXPLAIN ANALYZE SELECT * FROM generate_series(100);
+-------------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| plan_type         | plan                                                                                                                                                                                             |
+-------------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Plan with Metrics | LazyMemoryExec: partitions=1, batch_generators=[generate_series: start=0, end=100, batch_size=8192], metrics=[output_rows=101, elapsed_compute=74.042µs, output_bytes=64.0 KB, output_batches=1] |
|                   |                                                                                                                                                                                                  |
+-------------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1 row(s) fetched.
Elapsed 0.006 seconds.

We can add a special marker to sqllogictest to skip those non-deterministic parts.

What changes are included in this PR?

• Changed sqllogictest validator to recognize <slt:ignore> marker
• doc
• slt test

Are these changes tested?

Are there any user-facing changes?

[martin-g]

This early return would make it hard to debug. It would be nice to log a warning with the reason, as done below (line 116+).

[2010YOUY01]

Honestly, I don't know what those log are used for. I think the error message is generated elsewhere, and I want to make sure they're developer-friendly.

It looks okay now, but with false positives

2. query result mismatch:
[SQL] select * from generate_series(3);
[Diff] (-expected|+actual)
-   <slt:ignore>
+   0
    1
-   3<slt:ignore>
+   2
    3
at /Users/yongting/Code/datafusion/datafusion/sqllogictest/test_files/slt_features.slt:84

I have filed #18878 for improvement.

[martin-g]

I think this logic won't detect a prefix in the actual_snapshot.
Let's say the first expected fragment is "Hello" but the actual_snapshot starts with "Anything here before Hello". The find() will set the position to the correct index but Anything here before will be ignored and assumed as matching.
Similar issue with a suffix in actual_snapshot.

[2010YOUY01]

This is a good catch. I fixed it in a3154e7 and added more tests.