fix(BA-6914): give the three drifting FKs one name on both install paths

[jopemachine]

📚 Stacked PRs

• fix(BA-6913): reconcile indexes that only exist on migrated databases #12907 — BA-6913 fix: indexes that only exist on migrated databases
• fix(BA-6914): give the three drifting FKs one name on both install paths #12910 — BA-6914 fix: one FK constraint name on both install paths ← you are here
• fix(BA-6915): drop the column default before retyping sessions.result #12912 — BA-6915 fix: drop the column default before retyping sessions.result

Merge in order, bottom-up. Each PR is based on the one above it, so its diff shows only its own layer.

Problem

Three migrations create a foreign key under a hardcoded name; the models left it unnamed and let the metadata naming convention generate one. Same key, same enforcement — different name depending on how the DB was built:

	migration creates	fresh install has
cd067180a8b1	fk_kernels_image_id	fk_kernels_image_id_images
c7f2a8e31b04	fk_login_sessions_login_client_type_id	fk_login_sessions_login_client_type_id_login_client_types
a3b4c5d6e7f8	fk_prometheus_query_presets_category_id	fk_prometheus_query_presets_category_id_prometheus_quer_8a51 ¹

¹ the convention name exceeds PostgreSQL's 63-char identifier limit, so SQLAlchemy truncates it and appends a hash.

Not cosmetic: each downgrade drops the constraint by the hardcoded name and aborts with UndefinedObjectError on a fresh install. Whether a downgrade works depends on where the DB came from.

Which name is correct?

The convention (fk_%(table_name)s_%(column_0_name)s_%(referred_table_name)s) is the house style — 81 of 82 model FKs are unnamed. By that count the hardcoded three are the deviation, which argues for renaming migrated DBs to the convention names.

This PR goes the other way, for three reasons:

1. It would break what works today. Those downgrades run fine on migrated (production) databases right now. Renaming to the convention would break them there too, forcing edits to three released migrations.
2. The convention name is unusable for one of them — truncated and hash-suffixed. Converging on it would hardcode ..._8a51, tying us to SQLAlchemy's truncation algorithm.
3. Explicit naming already exists here — fk_deployment_policies_endpoint in deployment_policy/row.py.

So: name the three explicitly on the models, and rename on the databases built from the convention. Nothing outside their own migration files references these names.

This is the reviewable decision in this PR — if you'd rather converge on the convention, the cost is editing those three downgrades plus hardcoding the truncated name.

Fix

1. name= on the three model FKs, matching what the migrations create and their downgrades drop.
2. c8d1e5b73f92 renames the constraint where it differs. It looks the current name up via the inspector rather than hardcoding the convention name — that keeps it independent of SQLAlchemy's truncation and makes it a no-op where the name already matches. Downgrade is a no-op: undoing it would only put the drift back, and the convention name is not reconstructible.

Verification

Real databases throughout; the dev DB was only ever read.

Fresh install with the fixed models — all three come out with the short names.

Existing fresh install (schema built from main's models, so convention names), after upgrade head:

table	before	after
kernels	fk_kernels_image_id_images	fk_kernels_image_id
login_sessions	fk_login_sessions_login_client_type_id_login_client_types	fk_login_sessions_login_client_type_id
prometheus_query_presets	fk_prometheus_query_presets_category_id_prometheus_quer_8a51	fk_prometheus_query_presets_category_id

The three walls, on a fresh install (stamp the revision, run only its downgrade) — all now pass:

migration	before	this PR
a3b4c5d6e7f8	FAILED — constraint "fk_prometheus_query_presets_category_id" does not exist	OK
cd067180a8b1	FAILED — constraint "fk_kernels_image_id" ... does not exist	OK
c7f2a8e31b04	FAILED — constraint "fk_login_sessions_login_client_type_id" ... does not exist	OK

Found while surveying the downgrade chain for #12881.

🤖 Generated with Claude Code