[dajiaohuang]

Problem

The documented subset(ctd, indices = ...) path compares the length of every data/flag vector with the number of selected indices, rather than the number of observations in the input. Selecting two rows from a five-row CTD therefore silently returns all five rows (and their original flags).

library(oce)
x <- as.ctd(c(34,34.2,34.4,34.6,34.8), c(12,11,10,9,8), c(0,5,10,15,20))
subset(x, indices=c(1L,3L))[["pressure"]]
# before: 0 5 10 15 20
# after:  0 10

Change

Use the input temperature-vector length to identify observation-length data and flags. This retains the existing treatment of uncommon-length items from #2250. Add synthetic public-API regression coverage for shorter, single-row, reordered, negative and empty selections, flags and station metadata.

Validation

• R 4.3.3 / Ubuntu 24.04: source build, install and load passed.
• The public-API reproduction failed on develop 131d14e99fe5ad3fbce68823ca33a28291fffe25 and passes with this change; new testthat file passes.
• Synthetic CTD conversion, logical subset, metadata, a small synthetic SBE CNV read, selected numeric APIs, headless PDF profile plotting and the reviewed synthetic as.ctd example passed.
• R CMD check --no-tests --no-examples --no-vignettes --no-manual: completed with 2 warnings (unbuilt vignettes) and 3 notes (unavailable suggests/xrefs and installed size). Full upstream tests/examples and vignette execution were not run; no private observation files, external data services or instruments were used.

[richardsc]

Tests pass, I'll merge.