fix(BA-1840): Make cloud provider detection future-proof

[achimnol]

resolves #5085 (BA-1840)

This PR modernizes ai.backend.common.networking.curl() and ai.backend.common.identity.{detect_cloud(), get_instance_*()} functions.

• Use the officially documented metadata URLs for each cloud provider instead of relying on hacky sysfs-based DMI vendor detection to make it future-proof.
  • Tested actual working on Azure (DSv5), AWS (t3), and a local Linux VM.
• Update the Azure/GCP instance ID generation logic to combine the VM name and BASE-32 encoded value of the last 5 bytes of VM UUID (e.g., i-my-vm-name-mfqwcylb) as the VM name is only guaranteed to be unique within resource groups where resource groups are just tags.
• Use asyncio.staggered.staggered_race() to run the happy-eyeballs logic over multiple cloud-provider detection routines.
  • NOTE: You have to re-raise (or "skip catching") exceptions if you want to indicate them as explicit suspension of each coroutine.
• Use aiohttp.ClientTimeout(connect=...) instead of async with asyncio.timeout() nor classic timeout=... parameter to explicitly set timeouts on the connection attempts only, excluding response parsing and processing.
• Use raise_for_status=True and catch aiohttp.ClientError which includes aiohttp.ClientResponseError.
  • Update test_curl_* test cases to use aioresponses to make raise_for_status=True option working in tests.
• Ensure importing ai.backend.common.identity module working regardless whether there is a running event loop or not.
• Avoid using assert and handling AssertionError for mandated runtime checks as they may be omitted with Python runtime's optimization flags.

Example

AWS

current_provider = amazon
instance_id      = i-010a01d9e13073306
instance_ip      = 172.26.8.181
instance_type    = t3.xlarge
instance_region  = amazon/ap-northeast-2

Azure

current_provider = azure
instance_id      = i-base-image-test-p3l253at
instance_ip      = 10.130.0.5
instance_type    = Standard_D2ads_v5
instance_region  = azure/KoreaCentral

GCP

current_provider = google
instance_id      = i-instance-20250713-054151-663n56lh
instance_ip      = 10.132.0.2
instance_type    = e2-highcpu-8
instance_region  = google/asia-northeast3-c

Local VM

current_provider = None
instance_id      = i-localdev
instance_ip      = 127.0.1.1
instance_type    = default
instance_region  = local

Checklist: (if applicable)

• Milestone metadata specifying the target backport version
• Mention to the original issue

[Copilot]

Pull Request Overview

This PR refactors cloud provider detection to be async and future-proof, replacing local heuristics with concurrent metadata endpoint probes and standardizing return values using a new CloudProvider enum.

• Introduces CloudProvider enum and per-provider async detection helpers
• Replaces synchronous detect_cloud with asyncio.staggered_race for fastest metadata lookup
• Enhances curl helper with aiohttp timeouts, raise_for_status, and type-safe overloads

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
src/ai/backend/common/networking.py	Updated curl to use aiohttp.ClientTimeout, added overloads
src/ai/backend/common/identity.py	Rewrote detect_cloud with async sub-detectors, added CloudProvider enum and asyncio.staggered_race

Comments suppressed due to low confidence (2)

src/ai/backend/common/identity.py:97

• [nitpick] Consider adding unit tests or integration tests for detect_cloud to validate correct provider detection and fallback behavior across AWS, Azure, and GCP.

async def detect_cloud() -> Optional[CloudProvider]:

src/ai/backend/common/networking.py:23

• The runtime annotation references yarl.URL but yarl is only imported under TYPE_CHECKING. Either import yarl at runtime or enable postponed annotations (from __future__ import annotations) to avoid a NameError.

    url: str | yarl.URL,