docs: add confidential model serving guide

## Summary

KServe v0.20 introduced confidential model serving, enabling encrypted ML model artifacts to be decrypted only inside Trusted Execution Environments (TEEs) using hardware attestation. No documentation exists for this feature.

## What needs to be documented

- Overview of confidential model serving and its use cases (protecting model IP)
- How it works: storage initializer → JWE detection → CDH contact → KBS attestation → in-TEE decryption
- `confidential` spec field on both InferenceService (`PredictorExtensionSpec`) and LLMInferenceService (`LLMModelSpec`)
- `enabled: true` and `resourceId` fields (`kbs://///` format)
- RATS-protocol agnostic design (works with Trustee, Intel Trust Authority, etc.)
- Prerequisites: CoCo (Confidential Containers), TEE-capable nodes (SEV-SNP, TDX, etc.)
- Example YAML for both InferenceService and LLMInferenceService
- Validation rules: requires `storageUri` when enabled, warns on OCI URIs

## Related PR

- https://github.com/kserve/kserve/pull/5382

## Since

KServe v0.20

Start with related PR #5382 and KServe v0.20's confidential model serving changes. Verify the documented fields, prerequisites, flow, validation rules, and YAML examples for both InferenceService and LLMInferenceService. Done means the guide covers the listed behavior accurately and explains the required storageUri and resourceId settings.

Written by the indexing model from the issue text.