Fix crash when recording a failed execution for a job that already has one

The previous code used `create_or_find_by!` in `Job#failed_with`, passing
an `exception` object as an attribute. When a FailedExecution already
existed for the job (due to a unique constraint violation), the
`find_by!` fallback tried to use the exception object as a SQL bind
parameter, causing `TypeError: can't cast ProcessMissingError`.

This state — a job having both a ClaimedExecution and a FailedExecution
— shouldn't be possible given the transactional guarantees in
`ClaimedExecution#failed_with`, but has been observed in practice by
multiple users. When it happens, `fail_orphaned_executions` crashes on
startup, preventing Solid Queue from starting at all.

Replace `create_or_find_by!` with `create!`, rescuing RecordNotUnique
to find and update the existing FailedExecution with the new error
details. If the record disappears between the failed create and the
find (race with a concurrent retry), retry the create.

Also change the `expand_error_details_from_exception` callback from
`before_create` to `before_save` (guarded by `if: :exception`) so that
updating an existing FailedExecution with a new exception properly
serializes the error details.

Fixes #699

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: rosa

app/models/solid_queue/failed_execution.rb

@@ -6,7 +6,7 @@ class FailedExecution < Execution
 
     serialize :error, coder: JSON
 
-    before_create :expand_error_details_from_exception
+    before_save :expand_error_details_from_exception, if: :exception
 
     attr_accessor :exception
 

app/models/solid_queue/job/retryable.rb

@@ -16,7 +16,14 @@ def retry
       end
 
       def failed_with(exception)
-        FailedExecution.create_or_find_by!(job_id: id, exception: exception)
+        FailedExecution.create!(job_id: id, exception: exception)
+      rescue ActiveRecord::RecordNotUnique
+        if (failed_execution = FailedExecution.find_by(job_id: id))
+          failed_execution.exception = exception
+          failed_execution.save!
+        else
+          retry
+        end
       end
 
       def reset_execution_counters

test/models/solid_queue/claimed_execution_test.rb

@@ -73,6 +73,23 @@ class SolidQueue::ClaimedExecutionTest < ActiveSupport::TestCase
     assert job.reload.failed?
   end
 
+  test "fail with error when a failed execution already exists updates the existing one" do
+    claimed_execution = prepare_and_claim_job AddToBufferJob.perform_later(42)
+    job = claimed_execution.job
+
+    # Simulate corrupted state: a failed execution already exists for this job
+    SolidQueue::FailedExecution.create!(job_id: job.id, exception: RuntimeError.new("old error"))
+
+    assert_no_difference -> { SolidQueue::FailedExecution.count } do
+      assert_difference -> { SolidQueue::ClaimedExecution.count }, -1 do
+        claimed_execution.failed_with(RuntimeError.new("new error"))
+      end
+    end
+
+    assert job.reload.failed?
+    assert_equal "new error", job.failed_execution.message
+  end
+
   test "provider_job_id is available within job execution" do
     job = ProviderJobIdJob.perform_later
     claimed_execution = prepare_and_claim_job job