[Bug]: index.json files disappear when disk is full

[tendant]

zot version

v2.1.12-0-g331a927

Describe the bug

When disk becomes full (ENOSPC) during an image push, the repository index.json file is removed, causing all existing images in the repository to become inaccessible.

This leads to permanent loss of repository metadata under a common failure condition.

To reproduce

1.	Run zot with local storage (/data/zot)
2.	Fill disk until nearly full
3.	Push a new image to an existing repository
4.	Disk reaches 100% during push
5. 	index.json file under the repository directory is deleted / missing
6. 	Repository becomes empty

    "distSpecVersion": "1.1.1",
    "storage": {
        "rootDirectory": "/data/zot"
    },
    "http": {
        "address": "0.0.0.0",
        "port": "8000",
        "compat": ["docker2s2"],
        "auth": {
            "htpasswd": {
                "path": "/etc/zot/htpasswd"
            }
        }
    },
    "log": {
        "level": "debug"
    }
}```

### Expected behavior

	•	Existing index.json should remain intact if write fails
	•	Failed push should not destroy repository metadata

### Screenshots

_No response_

### Additional context

_No response_

[rchincha]

This would be unusual!

We have the following logs/behavior (line numbers below could be slightly off)

When disk is full (or any storage write/commit/move returns an unexpected error), the API path generally returns 500 Internal Server Error.

Write-failure logs in storage layer:

Manifest blob write failure logs failed to write in imagestore.go:648.
Monolithic blob upload write failure logs failed to write blob in imagestore.go:1187.
Monolithic blob upload commit failure logs failed to commit blob in imagestore.go:1195.
Finish upload commit failure logs failed to commit file in imagestore.go:1092.
Finish upload move failure logs failed to finish blob in imagestore.go:1144.
HTTP status mapping in handlers:

Blob upload PATCH/finish unexpected error branch:
Log: unexpected error, removing .uploads/ files in routes.go:1770 and routes.go:1803.
Response: 500 via WriteHeader(StatusInternalServerError) in routes.go:1776 and routes.go:1809.
Blob upload monolithic POST branch:
Log: failed to full blob upload in routes.go:1468.
Response: 500 in routes.go:1469.
Manifest PUT unexpected error branch:
Log: unexpected error, performing cleanup in routes.go:798.
Response: 500 in routes.go:808.
Net result for disk-full write failures:

Log(s): one or more of the storage write/commit/move errors above, plus an API-level unexpected error log.
HTTP status: 500 Internal Server Error.

[andaaron]

This would be unusual!

We have the following logs/behavior (line numbers below could be slightly off)

When disk is full (or any storage write/commit/move returns an unexpected error), the API path generally returns 500 Internal Server Error.

Write-failure logs in storage layer:

Manifest blob write failure logs failed to write in imagestore.go:648. Monolithic blob upload write failure logs failed to write blob in imagestore.go:1187. Monolithic blob upload commit failure logs failed to commit blob in imagestore.go:1195. Finish upload commit failure logs failed to commit file in imagestore.go:1092. Finish upload move failure logs failed to finish blob in imagestore.go:1144. HTTP status mapping in handlers:

Blob upload PATCH/finish unexpected error branch: Log: unexpected error, removing .uploads/ files in routes.go:1770 and routes.go:1803. Response: 500 via WriteHeader(StatusInternalServerError) in routes.go:1776 and routes.go:1809. Blob upload monolithic POST branch: Log: failed to full blob upload in routes.go:1468. Response: 500 in routes.go:1469. Manifest PUT unexpected error branch: Log: unexpected error, performing cleanup in routes.go:798. Response: 500 in routes.go:808. Net result for disk-full write failures:

Log(s): one or more of the storage write/commit/move errors above, plus an API-level unexpected error log. HTTP status: 500 Internal Server Error.

This analysis takes into considerations only blob upload related errors.
The error is on updating the index.json of the repo, which looses content regardless of status code returned by the API call.

[andaaron]

Merged this fix: #3963