pref: Optimize xrefmap download performance

[filzrev]

This PR is intended to resolve #9558

What's included in this PR

• Add settings to use HTTP Compression (if available)
• Change HttpClient's timeout to 30 minutes (Default: 100 seconds)
• Change StreamReader's buffer size to 80KB (Default: 1024 byte)
  • 80KB is default buffer size that used by Stream::CopyTo.

What's tested

I've manually tested xrefmap download with following configuration.

  "build": {
    "xref": [
      "https://learn.microsoft.com/en-us/dotnet/.xrefmap.json"
    ]
  }

And confirmed server returns following HTTP Response headers. And download size is reduced. (316MB -> 16MB)

HTTP/1.1 200 OK
Content-Type: application/json
Content-Length: 15978441
Content-Encoding: gzip

TODO Tasks
I'll create another PR later relating #9558.

• Add special deserialization code path for JSON file.
  (Because YamlDotNet based JsonDeserializer is 5x slower than System.Text.Json based deserialization)
• Add support for loading xrefmap compressed by gzip (xrefmap.json.gz) from local disk.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (ad6db16) 74.31% compared to head (7e13411) 74.30%.
Report is 7 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #9637      +/-   ##
==========================================
- Coverage   74.31%   74.30%   -0.01%     
==========================================
  Files         536      536              
  Lines       23173    23177       +4     
  Branches     4054     4054              
==========================================
+ Hits        17220    17221       +1     
- Misses       4851     4853       +2     
- Partials     1102     1103       +1     

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

[yufeih]

Looks great! Thank you @filzrev