Use atomic file writes for query cache

[spawnia]

@thekonz Turns out you were right all along to also use atomic writes for the query cache files in your original pull request.

After deploying the new query cache to production, I saw the following error a couple of times:

TypeError: GraphQL\Utils\AST::fromArray(): Argument #1 ($node) must be of type array, int given, called in /var/www/vendor/nuwave/lighthouse/src/Cache/QueryCache.php on line 152
#56 /vendor/webonyx/graphql-php/src/Utils/AST.php(74): GraphQL\Utils\AST::fromArray
#55 /vendor/nuwave/lighthouse/src/Cache/QueryCache.php(152): Nuwave\Lighthouse\Cache\QueryCache::requireOPcacheFile
#54 /vendor/nuwave/lighthouse/src/Cache/QueryCache.php(118): Nuwave\Lighthouse\Cache\QueryCache::fromHybridOrParse
#53 /vendor/nuwave/lighthouse/src/Cache/QueryCache.php(73): Nuwave\Lighthouse\Cache\QueryCache::fromCacheOrParse
#52 /vendor/nuwave/lighthouse/src/GraphQL.php(287): Nuwave\Lighthouse\GraphQL::parse
...

I believe it comes from concurrent writes to the file.

[spawnia]

Released with https://github.com/nuwave/lighthouse/releases/tag/v6.63.1.

[thekonz]

i think it is a partial file being read into opcache.

[spawnia]

Yeah, if the written file is empty or only contains <?php (no return), require returns 1. I suspect what happened is that:

• two processes A and B ran at the same time
• process A tries to write the file, the first step for that is that an empty file is created
• process B sees the empty file and tries to read from it

Now, this change makes sure that the file is filled once it exists at the proper path.