Fix: don't create .Random.seed as side effect of loading shiny

[gadenbuie]

Fixes #4382

Summary

Loading shiny in a fresh R session (where .Random.seed doesn't yet exist) caused .Random.seed to be created as a side effect. This happened because withPrivateSeed() unconditionally called httpuv::getRNGState() on exit, which calls R's C-level GetRNGstate() — and that function initializes the RNG (creating .Random.seed) when it doesn't already exist.

The fix moves the httpuv::getRNGState() call inside the branch that restores a pre-existing seed. When there was no .Random.seed before the call, we simply remove it and skip the sync — leaving the RNG in the same "not yet initialized" state it was in before package load.

Test plan

• Existing test-utils.R tests pass (seed save/restore behavior unchanged)
• Manually verify in a fresh R session:

  exists(".Random.seed", envir = globalenv(), inherits = FALSE)
  #> FALSE
  loadNamespace("shiny")
  exists(".Random.seed", envir = globalenv(), inherits = FALSE)
  #> FALSE