The #740 relock downgraded biolink-model, curies, linkml and koza to satisfy an exact pin — not forced by oaklib, and an upward path exists #741

Found reviewing #740.

uv lock --upgrade-package oaklib (0.6.23 → 0.7.4) also moved, downward:

Nothing in oaklib 0.7.4 bounds any of these (it declares curies>=0.6.6,
linkml-runtime>=1.5.3). The chain is: biolink-model 4.3.6 pins linkml==1.9.4,
linkml-runtime==1.9.5, curies>=0.9,<0.10 exactly; something in oaklib 0.7.4's
new tree conflicts with one of those, so the resolver — told to move only oaklib
and keep everything else as close to the lock as possible — fell back to
biolink-model 4.2.5, which pins curies<0.8, and that dragged curies, linkml and
koza down with it.

It is not forced. uv lock --dry-run --upgrade-package biolink-model --upgrade-package curies --upgrade-package linkml --upgrade-package koza on the
#740 lock resolves cleanly to biolink-model 4.4.4, curies 0.15, koza 2.6.2,
linkml 1.11.1 and linkml-runtime 1.11.1. That is the fix, and it is a
different PR: this repo is linkml-heavy (validators, gen-python, term/reference
validators), and 1.9 → 1.11 needs its own verification pass. #740 should not
carry it.

Why it is safe to ship #740 as is: nothing under src/, scripts/ or tests/
imports curies, koza or biolink directly (koza/biolink-model are only in the
unused koza extra); the full suite passes on the downgraded set; and the CLI,
linkml-validate, and linkml-term-validator were smoke-tested on it. But a
patch-level linkml downgrade and a two-minor curies downgrade sitting silently in
the lock is the kind of thing that gets blamed on the wrong change later, hence
the record.