Introduce parallelism for batch conversion: --parallel / -P

[yhatt]

This PR introduces --parallel / -P option to set the number of concurrency of parallel conversion for multiple files.

# Convert Markdown files in `./marp-slides` parallel (up to 5 workers: default)
marp --parallel 5 ./marp-slides

By applying the robust async process (async initialization is running only once, and all of successors will wait that initialization) to browser operations, the parallel processing could be achieved in a simple queue and workers.

Marp CLI will use up to 5 workers while converting multiple files in the queue by default. If used Marp CLI for converting multiple Markdowns into PDF/PPTX/image(s), this setting will make drastically faster conversion than previous. CLI user can tweak the number of concurrency by using --parallel or -P option.

Resolves #509.

Performance improvement

marp-parallel folder is including 30 empty Markdown files, and Marp CLI tries to convert every Markdown files into PDF.

CLI command	Parallelism	Total time	Faster	CPU usage
time marp --pdf --no-parallel ./marp-parallel	1	64.71 sec	-	18%
time marp --pdf --parallel=5 ./marp-parallel	5 (default)	14.50 sec	✨ x4.46	67%
time marp --pdf --parallel=10 ./marp-parallel	10	8.565 sec	✨ x7.56	108%

Increasing the number of concurrency from 5 to 10, the conversion will become much faster, but the CPU usage may stick to 100% while conversion in my environment. For the balance between the speed and CPU, I decided the default concurrecny as 5.

In PDF/PPTX/image conversion, that number is similar to the meaning of "max browser tabs for operating internal conversion concurrency". So I think the best default of this value may not mean the number of CPUs, that is known as better default setting in common concurrency operations. (Even if the computer was 2 core, it's very popular case to open 3 and more tabs in the browser; That is a reason why the default number of concurrency of parallel conversion is a fixed number)