Stub `ForkedStorageCheck.storage_available?` by default in all specs

While looking to see what I could do to speed up the spec/features/security/* specs, I ran them with bin/rspec-stackprof and noticed this:

gitlab-ce master % ES bin/rspec-stackprof spec/features/security/project/public_access_spec.rb

Finished in 2 minutes 6.4 seconds (files took 5.16 seconds to load)
306 examples, 0 failures

Randomized with seed 22643

==================================
  Mode: wall(1000)
  Samples: 107374 (12.97% miss rate)
  GC: 10300 (9.59%)
==================================
     TOTAL    (pct)     SAMPLES    (pct)     FRAME
     19766  (18.4%)       19718  (18.4%)     Gitlab::Git::Storage::ForkedStorageCheck#timeout_check
     10278   (9.6%)        5730   (5.3%)     ActionView::PathResolver#find_template_paths
      8443   (7.9%)        4217   (3.9%)     ActionDispatch::Journey::Formatter#missing_keys
      8590   (8.0%)        2956   (2.8%)     Sprockets::Cache::FileStore#get
      2811   (2.6%)        2530   (2.4%)     Peek::PGInstrumented#async_exec
      2757   (2.6%)        2289   (2.1%)     Peek::PGInstrumented#exec_prepared
      6627   (6.2%)        1828   (1.7%)     ActiveSupport::FileUpdateChecker#max_mtime
      1775   (1.7%)        1698   (1.6%)     GRPC::ActiveCall#request_response
      1629   (1.5%)        1629   (1.5%)     #<Module:0x007ffcc0812ee0>.distance
      1615   (1.5%)        1516   (1.4%)     GRPC::ActiveCall#remote_read
      1522   (1.4%)        1167   (1.1%)     Time#to_time
       969   (0.9%)         969   (0.9%)     URI::RFC3986_Parser#split
      1150   (1.1%)         935   (0.9%)     ActiveRecord::LazyAttributeHash#[]
      1016   (0.9%)         865   (0.8%)     ActionDispatch::Journey::Path::Pattern#match
       760   (0.7%)         760   (0.7%)     ThreadSafe::NonConcurrentCacheBackend#[]
       695   (0.6%)         695   (0.6%)     #<Module:0x007ffcc00a66c0>.load_with_autoloading
       608   (0.6%)         608   (0.6%)     ActiveSupport::PerThreadRegistry#instance
       587   (0.5%)         587   (0.5%)     Gon.public_method_name?
       793   (0.7%)         565   (0.5%)     ActiveSupport::FileUpdateChecker#watched
       560   (0.5%)         560   (0.5%)     Sprockets::PathUtils#stat

So I benchmarked the method with 250 iterations, and then compared it against a stub of the method that just returns true straight away:

n = 250

path = File.expand_path("../tmp/tests/repositories", __dir__)

Benchmark.bm(20) do |x|
  x.report("unstubbed") do
    n.times { Gitlab::Git::Storage::ForkedStorageCheck.storage_available?(path, 5) }
  end
  x.report("stubbed") do
    allow(Gitlab::Git::Storage::ForkedStorageCheck)
      .to receive(:storage_available?).and_return(true)
    n.times { Gitlab::Git::Storage::ForkedStorageCheck.storage_available?(path, 5) }
  end
end

#                           user     system      total        real
# unstubbed              0.410000   0.350000  90.710000 ( 93.097250)
# stubbed                0.010000   0.000000   0.010000 (  0.011941)

Obviously a pretty big difference.

This MR stubs that method globally in the test suite unless an example includes the check_storage_availability metadata.

Here's the "after" for that file:

Finished in 1 minute 39.1 seconds (files took 5.75 seconds to load)
306 examples, 0 failures

Randomized with seed 4500

==================================
  Mode: wall(1000)
  Samples: 81576 (16.17% miss rate)
  GC: 8437 (10.34%)
==================================
     TOTAL    (pct)     SAMPLES    (pct)     FRAME
     10197  (12.5%)        5662   (6.9%)     ActionView::PathResolver#find_template_paths
      8651  (10.6%)        4321   (5.3%)     ActionDispatch::Journey::Formatter#missing_keys
      7985   (9.8%)        2705   (3.3%)     Sprockets::Cache::FileStore#get
      2674   (3.3%)        2359   (2.9%)     Peek::PGInstrumented#async_exec
      2610   (3.2%)        2159   (2.6%)     Peek::PGInstrumented#exec_prepared
      6240   (7.6%)        1734   (2.1%)     ActiveSupport::FileUpdateChecker#max_mtime
      1710   (2.1%)        1637   (2.0%)     GRPC::ActiveCall#request_response
      1532   (1.9%)        1532   (1.9%)     #<Module:0x007fae2180aee0>.distance
      1575   (1.9%)        1500   (1.8%)     GRPC::ActiveCall#remote_read
      1406   (1.7%)        1054   (1.3%)     Time#to_time
       929   (1.1%)         929   (1.1%)     URI::RFC3986_Parser#split
       971   (1.2%)         829   (1.0%)     ActionDispatch::Journey::Path::Pattern#match
       983   (1.2%)         790   (1.0%)     ActiveRecord::LazyAttributeHash#[]
       692   (0.8%)         692   (0.8%)     #<Module:0x007fae208a66c0>.load_with_autoloading
       682   (0.8%)         682   (0.8%)     ThreadSafe::NonConcurrentCacheBackend#[]
       790   (1.0%)         566   (0.7%)     ActiveSupport::FileUpdateChecker#watched
       554   (0.7%)         554   (0.7%)     Sprockets::PathUtils#stat
       533   (0.7%)         533   (0.7%)     Gon.public_method_name?
      4147   (5.1%)         532   (0.7%)     Sprockets::Cache::FileStore#safe_open
       564   (0.7%)         529   (0.6%)     ActiveRecord::Inheritance::ClassMethods#base_class

This should provide a not-insignificant improvement to the speed of any test that uses an on-disk repository.

@reprazent What do you think about this? Should we be stubbing something further up the call chain, like disabling the Feature.enabled? check, which we globally set to true?