precompile: skip small functions

[vtjnash]

These are likely to never be called directly, and would be quick to compile if they happened to be. In v0.6, I used to be manually excluding many of these from the sysimg; this is just a more general.

before
2565541888 maximum resident set size # 2.5 GB

after
1980424192 maximum resident set size # 2.0 GB

(build time and .ji size are also about proportionally (~25%) improved)

[KristofferC]

Running with Pkg3 precompilation on, I get the following with GNU Time:

Before PR:

   Command being timed: "make"
        Elapsed (wall clock) time (h:mm:ss or m:ss): 6:43.50
        Maximum resident set size (kbytes): 3259596

After PR:

 Command being timed: "make"
        Elapsed (wall clock) time (h:mm:ss or m:ss): 5:59.28
        Maximum resident set size (kbytes): 2551840

There seem to be some slight regressions in first call timing 0.5s -> 0.8s for first call to status and 1.5s -> 1.8s for first call to add so not that bad.

[JeffBezanson]

Seems like a huge win to me.

[KristofferC]

FWIW, current master (no Pkg3 precompile) is at Maximum resident set size (kbytes): 2843524. I haven't tried this PR without Pkg3 precompile yet.

[staticfloat]

Verification runs on buildbots:

ubuntu32: https://build.julialang.org/#/builders/68/builds/817 (this one doesn't really matter, it's just as verification that failures on x86 aren't due to something weird about the old CentOS images we use for the linux32 builder)
linux32: https://build.julialang.org/#/builders/47/builds/811
win32: https://build.julialang.org/#/builders/44/builds/819

[staticfloat]

Confirmed that this PR passes tests on all the buildbots, tragically the win32 buildbot fails on make docs after passing all tests.

[JeffBezanson]

I don't see any error messages from the doc build. Is there a way to get them?

[staticfloat]

For some reason it's saved into docbuild.log instead of just written out, but it's TLS issues while trying to clone METADATA.jl.

[staticfloat]

I applied the hotfix to the windows buildbots, I'm re-running the win32 build of this branch here: https://build.julialang.org/#/builders/44/builds/822

[staticfloat]

We're good to go on win32.

[StefanKarpinski]

Based on the analysis @KristofferC did in #26767, the premise of this PR seems like it may be false.