In distributed setups (e.g. behind a load balancer or using blue-green deployments), we're experiencing issues with inconsistent asset hashes caused by the use of filemtime() in the AssetManager.

The problem occurs because the generatePath() method uses filemtime() to build the asset directory hash. On multiple nodes, file modification times may differ despite identical content, leading to mismatched hashes. This causes 404 errors or missing assets when users switch between nodes.

Sticky sessions can partially mitigate this in load-balanced environments, but blue-green deployments still cause hash mismatches when switching traffic from one environment to another.

We have a few questions:

1. What was the rationale behind using filemtime() as part of the hash?
2. Since the asset versioning already includes globalAssetsVersion (which is automatically incremented when clearing assets via the admin panel), is filemtime() still necessary?
3. If we override the behavior to exclude filemtime(), what are the risks?

This behavior makes it hard to maintain reliable deployments in production environments.