netx: investigate refraction-networking/utls (2/n)

[bassosimone]

Following up from ooni/probe-engine#229. We want to determine:

• how to sync up with the latest version of Go (is it needed?)
• if we can successfully parrot the latest Firefox (so we can measure pretending to be it)

[kelmenhorst]

Further we want to determine:

• whether to use mimicked or randomized fingerprints
• mimicking
  • mimic TLS implementations that are popular enough to make censors risk collateral damage
  • requires high maintenance (high update frequency in browsers)
  • risk: (slightly) deviant fingerprints can be detected and blocked using blacklists
  • useful to counter whitelisting censors
• randomized fingerprints
  • use TLS fingerprints with ~randomly set (while likely) parameters for each connection, making it difficult to identify a client
  • requires lower maintenance
  • risk: can be filtered out due to low popularity (and low collateral damage) by whitelisting censors
  • useful to counter blacklisting censors (which seem to be more common (?))
• cycle through a list of fingerprints, until a not-blocked fingerprint is found
  • stateful censors detect this behavior

(considerations refer to Frolov, S., & Wustrow, E. (2019). The use of TLS in Censorship Circumvention. In NDSS.)

[bassosimone]

Will continue the work in #1702