[Llama3-text vLLM integration] Modify Llama3 text model (new and old codebase) forward apis for vLLM compatibility

[skhorasganiTT]

Ticket

N/A

Problem description

• The forward call interface for the llama3 text models was incompatible with vLLM and had some minor bugs

What's changed

• Added generator_vllm.py::TtLlamaForCausalLM for vLLM model init and execution
• Added _easy_trace_text (handles capture trace and decode forward trace automatically for user) to LlamaGenerator and modified decode_forward_text (only valid decode entry point) to either call _easy_trace_text or _decode_forward_no_trace_text depending on the enable_trace arg.
• Added read_from_device arg to decode_forward_text so vLLM can perform async output processing during decode execution and modified _easy_trace_text and _decode_forward_no_trace_text to not read back outputs
• Made same modifications to old llama70b codebase to remove old codepaths in vLLM
• Modified llama_common.py::get_padded_prefill_len to pad to 128 if seq len < 128 since that is the min required for llama3 attention (same as currently done for llama3 demo and vision model)
• Note: this PR is required to enable Add support for TT Llama3 text models (1B,3B,8B,70B-new) vllm#48

Checklist

• Post commit CI passes
• Blackhole Post commit (if applicable)
• Model regression CI testing passes (if applicable)
• Device performance regression CI testing passes (if applicable)
• (For models and ops writers) Full new models tests passes
• New/Existing tests provide coverage for changes

[skhorasganiTT]

T3K unit/demo/freq tests: https://github.com/tenstorrent/tt-metal/actions/runs/12471128804
Single-card demo tests: https://github.com/tenstorrent/tt-metal/actions/runs/12471132832
TG unit/freq tests: https://github.com/tenstorrent/tt-metal/actions/runs/12471141537
TG demo tests: https://github.com/tenstorrent/tt-metal/actions/runs/12473148786
All post-commit tests: https://github.com/tenstorrent/tt-metal/actions/runs/12471164501