test: add release-based queue retry example (framework#1531)

[goravel-coder]

Summary

• Retry a failed job by releasing it back to the queue (attempt count preserved) instead of retrying in-process, powered by goravel/framework (PR #1531) and goravel/redis (PR #149); TestRetryable fails the first two attempts and succeeds on the third.
• Bump goravel/framework to c0b610b so the database queue driver preserves sub-second backoff delays (time.Time.Add) and the jobs-table migration stores reserved_at/available_at/created_at with millisecond precision (DateTimeTz(col, 3)); the broadcast backoff test now asserts real 100ms/200ms delays instead of whole-second stand-ins.
• Add retry_after to the database, redis1, and redis queue connections so a crashed worker's reservation expires and the job is recovered by other workers, and cover the retry-exhausted path via TestReleaseBasedRetryExhausted, which lands the job in failed_jobs once ShouldRetry gives up.

Why

Goravel's queue now supports release-based retries: when a job implements queue.JobWithShouldRetry and fails, it is released back to the queue with its attempt count preserved instead of being retried purely in-process, so retries survive worker restarts and can be picked up by any worker. The example repo consumes the framework changes that enable this, declares the retry_after reservation-expiry window required by the drivers, and demonstrates the flow with a job that fails the first two attempts and succeeds on the third — while also proving the exhausted path lands the job in failed_jobs.

Sub-second backoff now works end to end: the framework's database driver keeps the full delay (time.Time.Add) and the jobs table stores available_at with millisecond precision, so the earlier whole-second workaround was reverted and the tests exercise the real 100ms/200ms delays. The example migration matches the framework stub so the reference is copy-safe for production databases, and the self-pin replace directive for goravel/framework in go.mod is restored so the example consistently builds against the framework revision it demonstrates.

// app/jobs/test_retryable.go
package jobs

import (
	"errors"
	"time"
)

type TestRetryable struct{}

func (r *TestRetryable) Signature() string {
	return "test_retryable"
}

// Handle fails while the attempt count is within the failure window.
func (r *TestRetryable) Handle(args ...any) error {
	return errors.New("test retryable error")
}

// ShouldRetry releases the job back to the queue on failure with its attempt
// count preserved, so retries survive worker restarts. The 100ms delay is
// preserved by the database driver (millisecond precision).
func (r *TestRetryable) ShouldRetry(err error, attempt int) (bool, time.Duration) {
	if attempt <= 2 {
		return true, 100 * time.Millisecond
	}

	return false, 0
}