Snake casing WORD regex mistakingly splitting lowercase mixed-digit field names

Proto file:

```
message BadRegexTestMessage {
uint64 ds18b20_serial_code = 1;
}
```

Result:

```
class BadRegexTestMessage(betterproto.Message):
ds18_b20_serial_code: int = betterproto.uint64_field(1)
```

`ds18b20` becomes `ds18_b20`.

Use the provided Proto file and generated BadRegexTestMessage output as the reproduction case. Locate the WORD regex responsible for snake casing, run the generator with the ds18b20_serial_code field, and verify that the generated name remains ds18b20_serial_code rather than becoming ds18_b20_serial_code.

Written by the indexing model from the issue text.