chore(ci): explicit concurrency groups now the shared runner holds 2 slots

Capacity 1 on the ovh-main runner was the implicit mutex serializing every
workflow; sow-platform#189 raises it to 2. Side-effectful workflows get a
serialized group; PR-triggered checks get a per-ref group with cancel so a
superseded head frees its slot.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

.gitea/workflows/build-binaries.yml

@@ -11,6 +11,12 @@ permissions:
   code: read
   releases: write
 
+# The shared runner runs 2 jobs at once now (sow-platform#189); capacity 1 was
+# the implicit mutex. Serialize this workflow explicitly; never cancel mid-run.
+concurrency:
+  group: build-binaries
+  cancel-in-progress: false
+
 jobs:
   build-binaries:
     runs-on: nix-docker

.gitea/workflows/ci.yml

@@ -7,6 +7,12 @@ on:
 
 permissions: read-all
 
+# One run per ref: a new push obsoletes the run on the old head, so cancel it
+# and free the runner slot for the run that can still matter.
+concurrency:
+  group: ci-${{ gitea.ref }}
+  cancel-in-progress: true
+
 jobs:
   ci:
     runs-on: nix-docker

.gitea/workflows/sync-wrappers.yml

@@ -16,6 +16,12 @@ on:
 
 permissions: read-all
 
+# The shared runner runs 2 jobs at once now (sow-platform#189); capacity 1 was
+# the implicit mutex. Serialize this workflow explicitly; never cancel mid-run.
+concurrency:
+  group: sync-wrappers
+  cancel-in-progress: false
+
 jobs:
   sync:
     runs-on: nix-docker