query_habitat_thresholds_obs.R: availability km differ run to run in the last bits (float sum order) #293

Two runs of data-raw/query_habitat_thresholds_obs.R on the same database, commit and bundle differ in candidates.csv: avail_share_below is 0.11325447588492017 in one run and 0.11325447588492016 in the other, and ratio_below differs at the same digit. Measured 2026-09-26 while verifying #290.

The cause is sum(s.length_metre) / 1000 AS km (the availability query, ~line 371). It is a double-precision sum, and Postgres may parallelise the aggregate, so the summation order, and therefore the last bits, vary between runs. link's bar is byte-identical reruns, and this breaks it for a research producer whose outputs are committed as evidence.

Fix: sum as numeric (sum(s.length_metre::numeric) / 1000), or round to a fixed number of decimals before writing. Then re-run twice and cmp the outputs. The committed CSVs will change in the last digits once, so say so in the commit.

Relates to #284, #277