call `sync_with_transaction_state` inside `persisted?` then check ivars

directly

calling `sync_with_transaction_state` is not fast, so if we call it
once, we can improve the performance of the `persisted?` method.  This
is important because every call to `url_for(model)` will call
`persisted?`, so we want that to be fast.

Here is the benchmark:

```ruby
require 'active_record'

ActiveRecord::Base.establish_connection adapter: "sqlite3", database: ":memory:"

ActiveRecord::Base.connection.instance_eval do
  create_table(:articles)
end

class Article < ActiveRecord::Base; end
article = Article.new.tap(&:save!)

Benchmark.ips do |x|
  x.report("persisted?") do
    article.persisted?
  end
end
```

Before this patch:

```
$ bundle exec ruby -rbenchmark/ips persisted.rb
Calculating -------------------------------------
          persisted?     3.333k i/100ms
-------------------------------------------------
          persisted?     51.037k (± 8.2%) i/s -    253.308k
```

After:

```
$ bundle exec ruby -rbenchmark/ips persisted.rb
Calculating -------------------------------------
          persisted?     7.172k i/100ms
-------------------------------------------------
          persisted?    120.730k (± 5.1%) i/s -    602.448k
```

1 files changed, 2 insertions, 1 deletions

diff --git a/activerecord/lib/active_record/persistence.rb b/activerecord/lib/active_record/persistence.rb
index 7c076864a3..a6176dffdb 100644
--- a/activerecord/lib/active_record/persistence.rb
+++ b/activerecord/lib/active_record/persistence.rb
@@ -96,7 +96,8 @@ module ActiveRecord
     # Returns true if the record is persisted, i.e. it's not a new record and it was
     # not destroyed, otherwise returns false.
     def persisted?
-      !(new_record? || destroyed?)
+      sync_with_transaction_state
+      !(@new_record || @destroyed)
     end
 
     # Saves the model.