vtysh, lib: preprocess CLI graphs

[eqvinox]

Store a parsed and built graph of the CLI nodes in vtysh, rather than parsing and building that graph every time vtysh starts up.

This provides a 3x to 5x reduction in vtysh startup overhead:

vtysh -c 'configure' -c 'interface lo' -c 'do show version'

• before: 92.9M cycles, 1114 samples
• after: 16.5M cycles, 330 samples

This improvement is particularly visible for users scripting vtysh -c calls, which notably includes topotests.

[eqvinox]

I should make note of two things here:

• the speedup is even larger if you build vtysh with -no-pie. This is because the CLI graph serialized into vtysh contains a shitton of pointers, which turn into ELF relocations (300k of them to be exact) that the dynamic linker has to adjust at startup. This is absolutely showing up in perf data, but it's still much faster than dynamically parsing and creating the graphs at runtime.

• the few remaining DEFUN statements in vtysh (i.e. custom handlers) are still visible as significant overhead in perf data. Trying to somehow integrate those into preprocessing as well would yield another bit of gain, but it's much harder to do & this is where the cutoff is for now.

[eqvinox]

Report for command.h | 2 issues
===============================================
< ERROR: Macros with complex values should be enclosed in parentheses
< #362: FILE: /tmp/f1-4095093/command.h:362:
Report for vtysh.c | 2 issues
===============================================
< WARNING: externs should be avoided in .c files
< #1666: FILE: /tmp/f1-4095093/vtysh.c:1666:

These are junk warnings and will need to be ignored sigh

[eqvinox]

Looks like the remaining test failures are flaky timing. Faster vtysh will certainly highlight some problems with poorly written tests 😞

[donaldsharp]

ci:rerun

[donaldsharp]

I had this pushed in under #16501 closing this one