Fix boilerplate file resolution fallback to readonly data

[bwalderman]

Problem

Bikeshed 7.1.0 fails to build specs that rely on bundled boilerplate data, producing:

FATAL ERROR: Couldn't find an appropriate include file for the {name} inclusion, given Org 'w3c', Group 'webml', and Status 'CG-DRAFT'

This affects any spec without local boilerplate overrides. The same specs build successfully with 7.0.14.

Root Cause

Commit d924eca updated retrieveBoilerplateFile() in retrieve.py to search new-style paths (org-{org}/{group}/, org-{org}/, default/). The bundled readonly data in spec-data/readonly/boilerplate/ was correctly restructured to match.

However, retrieveBoilerplateFile() constructs FileInputSource objects using dataFile.path(), which only resolves paths in the latest (mutable) data directory (spec-data/boilerplate/). This directory either:

1. Doesn't exist on a fresh install (no bikeshed update has been run), or
2. Contains old-style paths (webml/, w3c/) after bikeshed update, because speced/bikeshed-data still serves the old flat structure.

In both cases, the new-style paths aren't found in the latest directory. The DataFileRequester fallback to readonly data is designed to work through fetch() (which catches OSError and retries with the fallback requester), but retrieveBoilerplateFile() bypasses fetch() entirely and it calls dataFile.path() to build a path string, wraps it in a FileInputSource, and calls .read() directly. The readonly fallback is never invoked.

How Found

CI for webmachinelearning/webmcp PR #204 started failing after the bikeshed 7.1.0 release. Traced the issue to the mismatch between the paths retrieveBoilerplateFile() searches and the paths that actually exist on disk.

Fix

Changed boilerplatePath() (singular) to boilerplatePaths() (plural), which returns candidate paths from both the latest and readonly data directories by consulting dataFile.fallback. For each boilerplate include, the search now tries the latest path first, then falls back to the readonly path restoring the intended fallback behavior that DataFileRequester provides through fetch().

This is a minimal, low-risk fix that doesn't change the search order or priority semantics. The underlying issue is that bikeshed update downloads boilerplate into old-style paths while the code expects new-style paths remains, but this fix ensures the bundled readonly data is always reachable as a fallback.

Testing

Verified against the webmachinelearning/webmcp spec (Status: CG-DRAFT, Group: webml).

[tabatkins]

Oh jeez, I know what the issue is. It's been a while since I last bumped the readonly files, so I forgot to increment the version.txt file as well; it's checked on every run and copies the readonly over into the active directory if readonly's version is higher.

Sorry, this is the sort of thing that's a little hard to catch during testing because my state is already fixed up to the latest version. Normally I use an internal bikeshed command to reset the readonly files, which tells me to bump the version number, but because this was limited to the boilerplates I hadn't run that.

I'm gonna go ahead and just bring the entire readonly directory up to date with the most recent bikeshed-data and then release 7.1.1 with the updated version; it'll pick things up properly. Sorry about that.

[tabatkins]

All right, 7.1.1 released with a fully updated spec-data/readonly directory and a bumped spec-data/readonly/version.txt, so it should trigger resetting your active spec-data files.