feat: improvements in package caching

[a-gave]

This adds some squid directives to consider fresh longer kernel modules of already compiled releases, to prevent revalidation against the origin server which consumes time.

An automatic mechanism to refresh the below squid directive should be included (e.g., 24h after the first appearance on https://downloads.openwrt.org/.versions.json)

refresh_pattern ^http://downloads.openwrt.org/releases/.*(23.05.[6-99]|24.10.[1-99]|25.*|SNAPSHOT)/targets/.* 0 100% 0 refresh-ims

This adds also a fallback to no_proxy (https) in case STDERR contains “package index are corrupt” to rule out cache problems and to allow packages to be considered fresh longer and thus not revalidated against the origin server

[codecov]

Codecov Report

Attention: Patch coverage is 52.38095% with 10 lines in your changes missing coverage. Please review.

Project coverage is 90.29%. Comparing base (5e65dec) to head (dc53a56).
Report is 225 commits behind head on main.

Files with missing lines	Patch %	Lines
asu/build.py	25.00%	6 Missing ⚠️
asu/util.py	69.23%	4 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1274      +/-   ##
==========================================
+ Coverage   80.75%   90.29%   +9.53%     
==========================================
  Files          15       13       -2     
  Lines         977     1205     +228     
==========================================
+ Hits          789     1088     +299     
+ Misses        188      117      -71     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[aparcar]

I recon you're really limited on bandwidth? Or what is the motivation to avoid the extra checks?

[a-gave]

Yes, this is purely to save bandwidth and to give a crutch to somewhat dated processors.

A build without selecting additional packages, just changing defaults

With a recent processor:
1- 43 seconds (refreshing everything)
2- 39 seconds (refreshing only packages, not kmods)
3- 22 seconds (without any refresh and with all TCP_MEM_HIT from ram)

The same with a 10years old processor:
1- 48s
2- 42s
3- 31s

Also the regex below should cause less fallback to https, if building with feeds from a mirror
(that resynchronizes i.e. every 12 or 24 hours)

refresh_pattern ^http://.*openwrt.*/.* 30 100% 30 override-expire

Close it if not necessary, maybe it could be useful as a reference for someone else, but maybe not.