ci: bound job runtime so a hung e2e cannot stall a release

The e2e suite talks to a real soulfind server over the network. On 298ec5a
it stopped responding and sat in_progress for over half an hour with every
other job green. No job declared timeout-minutes, so GitHub's six hour
default applied: six hours of runner time, and six hours during which the
new release gate correctly refuses to publish.

Limits are sized well above observed runtimes (lint ~1m, fmt ~15s, test
~2m, e2e ~1m) so they only fire on a genuine hang. A hang now surfaces as a
bounded red run that can be re-run, rather than an unbounded one.

This does not address why e2e hangs; it stops that from being unbounded.

.github/workflows/ci.yml

@@ -19,6 +19,7 @@ env:
 jobs:
   lint:
     name: Lint
+    timeout-minutes: 10
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v4
@@ -35,6 +36,7 @@ jobs:
 
   format:
     name: Format Check
+    timeout-minutes: 5
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v4
@@ -49,6 +51,7 @@ jobs:
 
   test:
     name: Test
+    timeout-minutes: 20
     runs-on: ${{ matrix.os }}
     strategy:
       matrix:
@@ -69,6 +72,7 @@ jobs:
 
   e2e:
     name: End-to-end (soulfind)
+    timeout-minutes: 15
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v4