Various improvements for test output

[matthewhughes934]

The main focus on this changeset is to reduce some noise that was
annoying me in test output (particularly in test failures)

There are still some compile errors outputted during tests (on failure
or with --nocapture due to doctests with compile_fail (there's an
open issue about this[1])

Link: rust-lang/rust#55813 [1]

• Update docs for Session.format

  To stop referencing parameters out and config that were both moved
  into the Session object with 71d3d04

• Move stray imports_granularity test

  Move this to be with all the other tests for this config option

• Silence expected parser errors

  I find otherwise the test output can be a bit noisy, and in particular
  the error messages make me think some test somewhere failed when even
  when it's all passing, for example, one such removed error:

    error: mismatched closing delimiter: `}`
     --> tests/parser/unclosed-delims/issue_4466.rs:3:17
      |
    2 |     if true {
      |             - closing delimiter possibly meant for this
    3 |         println!("answer: {}", a_func();
      |                 ^ unclosed delimiter
    4 |     } else {
      |     ^ mismatched closing delimiter
  

  This also removes some noisy warnings like:

    warning: whitespace symbol '\u{a0}' is not skipped
     --> tests/target/string_lit_unicode_ws.rs:3:22
      |
    3 |       let str = "hello \
      |  ______________________^
    4 | |  world";
      | | ^ whitespace symbol '\u{a0}' is not skipped
      | |_|
      |
  

• Remove warning on explicit default config value in tests

  This was added with f7a25a1, this was
  to address an issue with the worthy goal[1]:

  It is probably worth auditing the tests and where they are not
  deliberately testing an option, removing any options they set.

  However, given the number of warnings this produces:

  $ cargo test -- --nocapture |& grep -c 'Default value.*used explicitly'
  236
  

  I think this approach did not work, one could enforce this approach by
  changing this to be an error, but I think there are valid cases to use
  defaults, e.g. a reproduction of an issue relating to line length uses
  the default max_width: it's simplest to add a test verifying the
  behaviour with an explicit max_width: 100 rather than adjusting the
  reproduction.

  Link: improve test coverage for defaults #1720 [1]

[ytmimi]

r? @ytmimi

[rustbot]

This PR was rebased onto a different main commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.