fix(pinning): check before writing root address

[notanatol]

Checklist

• I have read the coding guide.
• I have filled out the description and linked the related issues.

Description

Closes: #4527

[ldeffenb]

I corrected my original review by moving that comment to Close() where it belongs, but added a new concern with the Has() check that was added to Put().

[ldeffenb]

This doesn't solve the problem of residual collectionChunks with their corresponding forever-remaining increments of the chunkstore RefCnt. Consider:

1. Two or more simultaneous operations start pinning the exact same content.
2. They all do the individual Puts, and the root collection doesn't yet exist, so they all actually store their collectionChunks and invoke chunkstore.Put incrementing RefCnt on those chunks.
3. One of the pinners actually executes the Close causing the pinCollectionItem to exist. That one "wins" and actually completes the pin.
4. When the other pinners finally invoke Close, they'll correctly not replace the original pinCollectionItem, but all of those incorrect RefCnt increments are still there causing those chunks to never be able to be removed from the chunkstore.

[notanatol]

Basically returning an error on line :276 will make cleanupOnErrWriter wrapper object to perform a cleanup (by calling the Cleanup method) whenever the error is detected

bee/pkg/api/api.go

Lines 870 to 874 in bf56d82

	func (r *cleanupOnErrWriter) WriteHeader(statusCode int) {
	// if there is an error status returned, cleanup.
	if statusCode >= http.StatusBadRequest {
	err := r.onErr()
	if err != nil {

notice the onErr pointing to the .Cleanup method

bee/pkg/api/bzz.go

Lines 106 to 110 in bf56d82

	ow := &cleanupOnErrWriter{
	ResponseWriter: w,
	onErr: putter.Cleanup,
	logger: logger,
	}

which in turn will decrement the ref count(s) to the correct value. See the test below.

[ldeffenb]

Thank you for that description. I should have followed the error path back out from Close() myself. I'm setting up a testnet environment to verify this on a full OSM tileset load whenever it is merged.

[ldeffenb]

Been tracking this one as it developed and the current version looks like it should solve the pinning refCnt issue that it intends to solve. Looking forward to testing it once it merges.