Fix n_rch_up/n_rch_down mismatch from flow corrections

[jameshgrn]

Bug

n_rch_up and n_rch_down scalar columns on 148 reaches did not match the actual neighbor counts in reach_topology. Root cause: flow correction flips updated topology direction entries (up/down) but did not update the scalar count columns.

93% (138/148) were exact swaps (n_rch_up had the correct n_rch_down value and vice versa). Affected regions: AF (10), AS (42), EU (14), NA (44), SA (38). OC had 0.

Impact

POM sword_validity.m Test 1b/2b detects this: declared n_rch_up count does not match actual upstream entries in rch_id_up array.

Fix

Single UPDATE recalculating from reach_topology. 148 reaches fixed. 0 remaining mismatches. All exports re-generated.

Separate: non-reciprocal pairs (151)

POM also flagged non-reciprocal topology (both reaches listing each other as upstream). Known v17b-inherited issue (documented in release notes as 150 pairs, actual count 151). Zero overlap with the n_rch_up/down bug.

Reported by: Pierre-Olivier Malaterre (INRAE)

[jameshgrn]

Also fixed the 151 non-reciprocal topology pairs (both reaches listing each other as upstream). These were v17c regressions from flow corrections, NOT inherited from v17b (v17b had 0 such pairs). Fixed by restoring v17b direction for the affected entries. Also updated n_rch_up/n_rch_down for the 151 affected reaches.

Global state after fix:

• Both-upstream pairs: 0
• n_rch_up/n_rch_down mismatches: 0

All exports being regenerated.

[jameshgrn]

Fixed in v17c-beta.0.0.1 — 148 mismatches corrected, 0 remaining across all 248,673 reaches. Exports re-generated.