use qwen3 1.7b as optional model

use tiiuae/Falcon-H1-1.5B-Deep-Instruct-GGUF:Q4_0 as model example

use default n_batch & n_ubatch

default to qwen3 0.6b

add -c option to force cpu only

use ernie 4.5 0.3b as default model

decrease batch number to save mem.

simplify script

add CLI interface

adjust output file name to summary.txt

output summary unquely in zh-TW

switch from tiny falcon 100m to ernie 4.5 21b a3b

cleanup: remove unused non-streaming summarize_transcript function

refactor: fix streaming summary bug and simplify to streaming-only output

summarization script

add llama-cpp-python JamePeng's fork