Change 'a' to 'an' [ci skip]

[ci skip] Minor improvements on routing.md.

I wanted the middle model to be anonymous, but we have to give it a name
for marshal support. :'(

This test was inspired by #14866. That PR would break this
functionality and we should make sure we notice that in the future.

Habtm relation size calculation fix. Fixes #14913, Fixes #14914.

This resulted in some unrelated errors on Traivs:

  1) Error:
PostgresqlUUIDTest#test_treat_blank_uuid_as_nil:
ActiveRecord::StatementInvalid: PG::UndefinedColumn: ERROR:  column "thingy" of relation "uuid_data_type" does not exist
LINE 1: INSERT INTO "uuid_data_type" ("thingy") VALUES ($1) RETURNIN...
                                      ^
: INSERT INTO "uuid_data_type" ("thingy") VALUES ($1) RETURNING "id"

/cc @zuhao

HABTM should fall back to using the normal CollectionAssociation's size calculation if the collection is not cached or loaded.

This addresses both #14913 and #14914 for master.

Mismatch branch name between text and code example [ci skip]

[ci skip]

Dup the changed_attributes otherwise we could lose them

Use Rails::Paths::Path#existent in database_configuration

Database configuration was trying to load the first path that
config.paths['config/database'] was returning even if the path didn't
exist in the filesystem.

Because Rails::Paths::Path has the possibility to return more than 1
path (as an array), database_configuration should filter down the paths
to the existing one and then load the first one.

This would make it possible to move the database.yml file and add the
new path to paths['config/database'] and still load the configurations.

Closes #14989

Fix Hash#deep_merge bug and improve documentation — resolves #12738

[related #13166]

[Regression 3.2 -> 4.0] Fix bugs with changed attributes tracking when transaction gets rollback

HTTP::Headers#key? correctly converts

Previously if you were looking for a given key, the header may incorrectly tell you that it did not exist even though it would return a valid value:

```ruby
env     = { "CONTENT_TYPE" => "text/plain" }
headers = ActionDispatch::Http::Headers.new(env)
headers["Content-Type"] 
# => "text/plain"

headers.key?("Content-Type")
# => false
```

This PR fixes that behavior by converting the key before checking for presence

This case prevents against regressions. The change was suggested in a recent
PR but the all our tests passed.

typo fixes in active_record_basics.md [skip ci]

WIP: Make ActionMailer tests order-independent.

rename var to fix overshadowing of variable

Follow-up to #14990 [ci skip]

Rewrite to avoid 'we'/'you', add missing period, and keep lines at 80 chars. Cheers :)

Improve readability with help from @senny

[ci skip]