version number fix - #467

self-explain

Author: Nan Zhu <[email protected]>

Closes #467 from CodingCat/branch-0.9 and squashes the following commits:

ba36109 [Nan Zhu] remove out-dated comments
9a8810e [Nan Zhu] version number fix

…only parent of the dist (apache#467)

Disable Orange jobs

… scale < 0 (apache#456)" (apache#467)

This reverts commit 0017da5

### What changes were proposed in this pull request?

Task 181's first owed item: the first-query cost, written into the closing post's plan (`PLAN_TASK_181.md` 9).

Task 195 measured the first query on 26 September, before task 212's warm-up existed, and found Varka slower than stock Spark at every rung: a new shape's kernel was never compiled over a hundred thousand rows. Task 212 then compiled new kernels in the background, on by default, and `VarkaColdStartBenchmark` gained the warm-up's arms. From that committed file (laptop, 100 000 Arrow-cached rows, best of five):

| entries | stock Spark, first run | Varka with warm-up, first run | stock Spark, second run | Varka, once compiled |
|---|---|---|---|---|
| 16 | 115 ms | 169 ms | 52 ms | 26 ms |
| 54 | 448 ms | 342 ms | 427 ms | 27 ms |
| 100 | 867 ms | 619 ms | 812 ms | 37 ms |

So bound 4 is restated: below the cliff Varka's first queries of a new shape cost about half again more; past it they already cost less; the cost is paid once per shape, since the shape cache keys a kernel by its tree and not its literals; and the warm-up's verdict takes about two seconds at 16 entries and twelve to thirteen at a hundred. The section gives 3.8 its paragraph in the post's voice, and records what is still owed: these are laptop numbers, and the post's numbers come from GitHub-hosted runners, so the benchmark runs there before publication (and will include task 221's warm-up fix, apache#467).

The milestone row notes the step.

### Why are the changes needed?

The outline listed the first-query cost as the one measurement standing between it and the draft; the measurement existed and its reading had changed.

### Does this PR introduce _any_ user-facing change?

No.

### How was this patch tested?

`dev/varka_quote_check.py` (every number traces to the committed results file) and the Varka pre-commit checks pass. Documentation only.

### Was this patch authored or co-authored using generative AI tooling?

Generated-by: Claude Code (Claude Opus 5.5)