feat(core): support offline validation and HTTP cache management

[kikkomep]

Summary

Introduce an offline validation mode that lets users validate RO-Crates without network access, relying on a persistent local HTTP cache for every resource the validator needs to resolve.

What changes

• Offline mode — a new option on the validator (exposed as --offline on the CLI) serves every HTTP-backed resource from a local cache, failing clearly when a resource has never been fetched.
• Persistent HTTP cache — the validator now uses a stable, user-level cache location (XDG-aware) so online and offline runs share the same store.
• Transparent caching of remote contexts — remote JSON-LD @context fetches performed by rdflib go through the same cache, which previously bypassed it.
• Cache pre-population (warm-up) — the validator can discover the external resources declared by the active profiles and prefetch them, both automatically (when online) and on demand.
• Cache management CLI — a new rocrate-validator cache command group lets users inspect, reset and pre-populate the cache.

User-facing changes

• validate --offline / validate --no-cache flags.
• new rocrate-validator cache info|reset|warm subcommand.

[fix #114 #115]

[elichad]

I've just tested this PR and it works well offline! The rocrate-validator cache commands look really helpful for our needs - will allow us to build the cache easily when we build our docker image before running it offline.

A few comments after my experiments - none are blocking, just quality of life stuff:

1. when I ran validation in offline mode against an un-cached profile, it emitted a useful warning telling me how to update the cache, but it emitted it a LOT of times (below I show just the last few instances, but there were many more) - could this be reduced so the error is only shown once?

 [2026-05-14 13:25:17,889] WARNING in models: Consider reporting this as a bug.                      
 [2026-05-14 13:25:17,928] WARNING in models: Unexpected error during check                             
 <rocrate_validator.requirements.shacl.checks.SHACLCheck object at 0x748a66147850>.  Exception: Resource                  
 'https://w3id.org/ro/crate/1.2/context' is not available in the HTTP cache and the validator is running in offline mode. 
 Run online once, or use `rocrate-validator cache warm` to pre-populate the cache.                                     
 [2026-05-14 13:25:17,928] WARNING in models: Consider reporting this as a bug.                      
 [2026-05-14 13:25:17,957] WARNING in models: Unexpected error during check                             
 <rocrate_validator.requirements.shacl.checks.SHACLCheck object at 0x748a6611cbd0>.  Exception: Resource                  
 'https://w3id.org/ro/crate/1.2/context' is not available in the HTTP cache and the validator is running in offline mode. 
 Run online once, or use `rocrate-validator cache warm` to pre-populate the cache.                                     
 [2026-05-14 13:25:17,957] WARNING in models: Consider reporting this as a bug.                      
 [2026-05-14 13:25:17,994] WARNING in requirements: Forced SHACL run for zero-shape target profile      
 five-safes-crate-0.4 failed: Resource 'https://w3id.org/ro/crate/1.2/context' is not available in the HTTP cache and the 
 validator is running in offline mode. Run online once, or use `rocrate-validator cache warm` to pre-populate the         
 cache.     

2. Normally I can run rocrate-validator validate <path> -p process-run-crate and it will find the process-run-crate-0.5 profile automatically. But if I run rocrate-validator cache warm -p process-run-crate I get the following output:

Profile(s) not found and skipped: process-run-crate
Nothing to warm up.

It would be useful if either cache warm could automatically find the versioned profile, or if it was more clearly documented that the full profile identifier must be used with cache warm.

3. I tried to manually cache https://w3id.org/ro/crate/1.2/context but couldn't achieve this. I tried:
   • rocrate-validator cache warm https://w3id.org/ro/crate/1.2/context (I don't think this is expected to work)
   • rocrate-validator cache warm --crate https://w3id.org/ro/crate/1.2 to hopefully cache the context from the RO-Crate 1.2 crate (error "Calling read(decode_content=False) is not supported after read(decode_content=True) was called")
   • rocrate-validator cache warm --crate https://www.researchobject.org/ro-crate/specification/1.2/ro-crate-metadata.json (same error)

It would be great if there was an easy way to provide a non-crate URL to cache.

[kikkomep]

Changes in the latest few commits:

1. Noisy repeated warning — fixed in 757b86a.

2. cache warm -p with non-fully-qualified profile — now automatically uses the latest profile version when a non-fully-qualified identifier is passed. A warning is emitted when more than one version is available, notifying the user that the latest is being selected automatically.

3. cache warm --url/-u — warm now supports caching of specific URLs via the new option.

Also added: cache list command (aliased ls) to list the cache contents.

@elichad let me know if it all looks good