Implements instance state filtering for both AutoScaling and tag-based
discovery modes. The `ec2:DescribeInstances` API calls now include
filters based on the `aws_ec2_instance_states` configuration.

The implementation adds `maybe_add_instance_state_filters/2` which
checks the configuration and conditionally applies state filters. When
states are configured, `add_instance_state_filters/3` builds a single
filter with multiple values in the format:

    Filter.N.Name=instance-state-name
    Filter.N.Value.1=running
    Filter.N.Value.2=pending

This format matches the AWS EC2 API specification and was verified with
the AWS CLI. The default configuration filters to `running` and
`pending` instances, excluding dead or dying instances (`stopping`,
`stopped`, `shutting-down`, `terminated`).

Both `get_hostname_by_instance_ids/2` (AutoScaling mode) and
`get_hostname_by_tags/1` (tag-based mode) apply the state filters
consistently. The filters are applied after tag filters and before the
final query string is built.

A unit test verifies the filter format is correct and matches the
expected AWS API parameter structure.

Adds tests for `get_hostname_by_instance_ids/2` and
`get_hostname_by_tags/1` to verify instance state filters are correctly
applied in both AutoScaling and tag-based discovery modes.

The tests use `meck` to mock `rabbitmq_aws:api_get_request/2`, avoiding
real EC2 API calls while verifying the query string format. Each test
confirms that:

- Instance state filters are present in the request path
- Filter format matches AWS API specification
- Tag filters and state filters work together correctly
- Hostname extraction from mock responses works as expected

The `get_hostname_by_tags_with_state_filter` test checks for
URL-encoded tag filters (`tag%3Aservice`) since the query string is
URL-encoded before the API call.

A helper function `mock_describe_instances_response/0` provides properly
formatted EC2 API response data for testing.

Adds validation for the `aws_ec2_instance_states` configuration to
ensure only valid EC2 instance state names are used. The validation
filters out invalid states and logs a warning, allowing the node to
start with the valid states.

The `validate_instance_states/1` function checks each configured state
against the list of valid EC2 instance states defined in the
`?VALID_EC2_INSTANCE_STATES` macro. Invalid states are discarded and
logged as a warning.

The `normalize_state/1` function handles both atom and string inputs,
converting atoms to strings for consistent handling. This supports
configuration via `advanced.config` with atoms (`[running, pending]`)
or via `rabbitmq.conf` with strings (`["running", "pending"]`).

Tests verify validation works correctly for:
- All valid states (strings)
- Mixed valid and invalid states (filters out invalid)
- Atom inputs (normalizes to strings)
- Mixed valid and invalid atoms