Add JarvisLabs provider

[peterschmidt85]

Adds JarvisLabs as an online gpuhunt provider.

Implementation notes:

• Fetches live offer data from JarvisLabs GET /misc/server_meta.
• Converts GPU rows into catalog items only when they are VM-capable (workload_type == "vm" or null). Container-only rows are ignored.
• Emits on-demand GPU offers from price_per_hour and spot GPU offers from spot_price when JarvisLabs reports spot capacity.
• Converts CPU VM offers from cpu_meta.combinations; CPU spot is not emitted because the API does not expose CPU spot pricing.
• Uses 100GB as the JarvisLabs VM minimum disk size and honors larger min_disk_size queries.
• Only emits offers for regions with known JarvisLabs VM provisioning hosts. server_meta does not include those hosts, so unknown VM regions are skipped with a warning.
• Skips malformed offer rows with warnings instead of failing the whole live catalog refresh.

E2E validation:

• dstack offer selection saw JarvisLabs L4 on-demand, L4 spot, and H100 spot offers from this provider.
• CPU on-demand, GPU on-demand, and GPU spot tasks were provisioned through dstack using these offers.

Added tests cover offer conversion, CPU-only offers, VM/container filtering, spot offers, unsupported-region filtering, malformed metadata, disk sizing, and catalog queries.