Small updates to allow for gpt-oss generation

[iamlemec]

Compatibility updates for gpt-oss generation:

• Seems like seq_rm doesn't accept -1 anymore, so just use seq_id = 0 for generation (is this fine for the server?)
• The gpt-oss template uses strftime_now in the template, so provide that function

Of course, this also needs a llama.cpp version bump (not included here) for the new model architecture.

[abetlen]

Thank you @iamlemec !