[killme2008]

Which issue does this PR close?

Related: GreptimeTeam#38

Rationale for this change

CASE expressions with literal WHEN values and nonliteral THEN expressions repeatedly copy unmatched rows, including string keys in partitioned join filters.

What changes are included in this PR?

Keep remaining row indices between branches and materialize selected rows for THEN/ELSE. Nonliteral WHEN expressions retain the existing path.

Performance measurements below use e6dfc8f8f9, including the nested-field-name compatibility fix.

Baseline 90b55e38b, M4 Max/48 GiB, Rust 1.99.1, release-nonlto; 8,192 rows and two 32-byte string keys. Criterion: ABA/BAB/ABA, 20 samples, 0.5s warmup and 1s measurement per block; three-round means:

CASE distribution	Baseline	This PR	Speedup
Uniform, 1 branch	16.669 µs	16.722 µs	0.997×
Uniform, 4 branches	225.847 µs	152.109 µs	1.485×
Uniform, 16 branches	418.673 µs	229.322 µs	1.826×
First branch only, 4 branches	16.700 µs	16.690 µs	1.001×
First branch only, 16 branches	16.777 µs	16.789 µs	0.999×
1 WHEN + ELSE, 50% match	138.987 µs	135.901 µs	1.023×

Uniform 4/16-branch maximum bracket drift was 3.9%. The 50% match case measured 1.041×/1.002×/1.025× across rounds, with maximum drift 2.4%. The mean change in the 50% match case is smaller than its maximum bracket drift.

The existing integer-column CASE measured 1.087× with maximum drift 5.1%. Unchanged lookup/no-base controls ranged from 1.000× to 1.005×; their maximum bracket drift was 11.6%.

What is the testing strategy for this PR?

Physical-expression tests cover row mapping, duplicate/NULL WHEN branches, error short-circuiting, zero-column batches and differing nested field names. The benchmark checks output before timing. Extended workspace tests (13,529 passed, 8 ignored), 546 SQLLogicTest files, clippy and formatting pass.

Are there any user-facing changes?

Performance only; SQL semantics and APIs are unchanged.

[killme2008]

@pepijnve @rluvaton, would you have time to review this CASE evaluation change?

It keeps row indices between literal WHEN branches to avoid repeatedly copying unmatched rows. Given your work on CASE evaluation, I'd appreciate your feedback. Thanks!