Add .protonignore and .gitignore support for keeping files out of sync

[Musiker15]

What this adds

Two ignore files, so that files can be kept out of synchronization the way .gitignore keeps them out of a repository.

.protonignore on the root of a sync folder applies to that whole sync root. .gitignore in any directory applies to that directory and everything below it, exactly where Git would apply it. Precedence is shallower .gitignore, then deeper .gitignore, then .protonignore, so a negated pattern in .protonignore can bring back something a repository .gitignore excludes.

The pattern syntax is the gitignore one with two Windows adjustments: matching is case insensitive, and both separators are accepted, so the backslash only escapes a leading # or !.

Where it hooks in

ItemExclusionFilter already existed for temporary files and special folders. It now returns a decision instead of a bool, so update detection can tell a built-in exclusion apart from a user rule. All six call sites in the enumeration and event log steps are unchanged, SuccessStep resolves the decision internally.

Rules are evaluated on the local replica only. The remote adapter gets NullIgnoreRuleProvider, because excluding a remote item would report it as gone and delete the local counterpart.

The part worth reviewing

An adapter that stops reporting an item is indistinguishable from a user deleting it, so a filter in this position is a deletion primitive. Two situations have to be told apart:

• The rule set changed and now matches something already indexed. The rule is skipped and the item keeps syncing, otherwise enabling a rule would delete data from every other device. This is logged.
• The item was renamed into an ignored name. Treated like any other exclusion, so the copy under the old name goes away. That is what the built-in exclusions have always done for renames into temporary file names.

They are told apart by comparing the indexed name with the incoming name.

The distinction is not cosmetic. Without it the feature barely works, because Explorer does not create anything under its final name. A new folder is "Neuer Ordner" first and a new file is "New Text Document.txt" first, so the item gets indexed under the placeholder name, then renamed into the ignored name, and is then pinned into synchronization for good. I hit exactly that in testing before adding the name comparison.

Verification

The matcher was checked against 35 cases covering anchors, negation, directory-only patterns, **, character classes, nested rule set chains and precedence.

Then a full run in a Windows VM against a test account with an empty sync folder. Eleven of eleven fixture cases behaved as specified, nested .gitignore files were confirmed to be loaded and applied to the right subtree, and both branches of the safety rule were confirmed in the log, for a file and for a directory.

PROTONIGNORE.md documents the syntax, the precedence and both directions of the safety behaviour.

Notes

I am aware the README says contributions are not accepted. Opening this anyway so that it is visible to anyone who wants the feature.

Unrelated to this change, the repository does not build from a clean clone. Proton.Cryptography and Proton.Drive.Sdk are not on nuget.org and no feed is configured, and src/Proton.Drive.Update/Releases/ is missing because [Rr]eleases/ in .gitignore matches it. See #1 and my comment there. My fork works around both, written up in BUILDING.md: https://github.com/Musiker15/windows-drive/blob/feature/ignore-rules/BUILDING.md