[Bug] The configurations of `second-to-micro` and `millis-to-micro` within `sequence.auto-padding` does not meet the expected behavior.

### Search before asking

- [X] I searched in the [issues](https://github.com/apache/incubator-paimon/issues) and found nothing similar.

### Paimon version

<=master

### Compute Engine

flink <=master

### Minimal reproduce step

The configurations of `second-to-micro` and `millis-to-micro` within `sequence.auto-padding` can lead to issues with nanotime generation (similar to data overflow). This can result in a combined sequence in later incoming data being smaller than that in historical data when they share the same `sequence.field`. Consequently, this might lead to the data being discarded, even though, theoretically, incoming data should always be the latest, provided they share the same `sequence.field`.

### What doesn't meet your expectations?

See https://github.com/apache/incubator-paimon/pull/1247#issuecomment-1670885102.

### Anything else?

_No response_

### Are you willing to submit a PR?

- [X] I'm willing to submit a PR!

Start by reviewing the sequence.auto-padding configuration and the linked PR comment describing the second-to-micro and millis-to-micro behavior. Reproduce the nanotime or overflow case and determine the expected ordering for incoming and historical records sharing a sequence.field; done means incoming records are not discarded because of a smaller combined sequence.

Written by the indexing model from the issue text.