Delay shutdown until the last sqs_message has been posted to the executor.

### Describe the feature

Hello, thanks for developing a great gem.
I have a request regarding the behaviour when SIGTERM fires.
My understanding is that the flow is currently as follows.
1. receive a SIGTERM
2. execute `raise Interrupt` with `trap`.
3. executor terminates as soon as the execution of the job in the queue is completed (until the timeout comes).

If `@executor.execute(sqs_message)` is executed on receipt of a SIGTERM, but there is a sqs_message that has not yet entered the queue because, for example, the max_queue has been reached, this message is not processed and is terminated.

We want the processing of this sqs_message to also continue until the timeout arrives.
Could you please consider changing the behaviour or adding this as an option?

### Use Case

We use a lot of Fargate spots and would like to increase the number of jobs that can be processed without waiting for a visibility timeout as much as possible from a performance perspective.

### Proposed Solution

Modified behaviour of Signal.trap.

1. set an instance variable to indicate that a SIGTERM signal has been received
2. `raise Interrupt` after executing execute_messages with `_poll`.

If you want to take into account the risk that the timeout setting will not function properly:
1. add another timeout during the waiting period of the last execute_messages

### Other Information

_No response_

### Acknowledgements

- [x] I may be able to implement this feature request
- [ ] This feature might incur a breaking change

### SDK version used

1.0.1

### Environment details (OS name and version, etc.)

Linux

Start by tracing Signal.trap through execute_messages, _poll, and @executor.execute(sqs_message) to understand when shutdown interrupts polling and queued work. Compare the current behavior with the proposed delayed interrupt, including the timeout while the final messages are submitted; done means messages received before shutdown are submitted and allowed to run without breaking timeout handling.

Written by the indexing model from the issue text.