Improve startup performance

[gsingh93]

Startup performance is not great, particularly when debugging a kernel with QEMU.

Some preliminary profiling with cProfile shows that syntax highlighting is a major issue, with set syntax-highlight off leading to a 43% improvement.

gdb-pt-dump also takes a significant amount of time, but there's work for a Rust rewrite underway, and in the meantime I plan to add an option to completely disable getting memory maps in QEMU to improve performance.

We should investigate these, and other issues, and start tracking startup time as part of every PR.

Edit: Seems like this call to print_help accounts for 8% of startup time:

pwndbg/pwndbg/commands/__init__.py

Line 360 in 41c4070

self.parser.print_help(file)

[disconnect3d]

Pretty sure the memoization improvements from #1453 may help here, at least a little bit.

[gsingh93]

Leaving a link to this here as something we can explore for performance benchmarking: https://pytest-benchmark.readthedocs.io/en/latest/

[disconnect3d]

Two things to profile this:

1. PYTHONPROFILEIMPORTTIME=1 gdb --quiet --batch - and checking import times (thx to https://stackoverflow.com/questions/34755728/profile-python-import-times/55603444#55603444)
2. PWNDBG_PROFILE=1 time gdb --quiet --batch && ./profiling/print_stats.py ./pwndbg-load.pstats > x && head -n50 x

Some quick observations:

• loading pwndbg on a low resource VPS takes 0.8-1s
• importing pwnlib is heavy
• importing requests - from ai.py is heavy

We should at least:

• defer import of requests to when it is actually needed in ai.py
• import only the used functionalities from pwnlib, not the whole module -- this may or may not help

[disconnect3d]

Some work has been done in:

• Optimize pwndbg.exception import time #1983
• Optimize pwndbg.commands.ai import time #1984

[disconnect3d]

Fwiw we should investigate "freezing" or "frozen" modules, since this may improve startup time if we would do it for ALL Pwndbg code and its dependencies code.

We could do it for Pwndbg releases?