Add a draft THREAT_MODEL.md + SECURITY.md and link it from AGENTS.md

[potiuk]

This is a v1 draft threat model for the OpenDAL PMC to review — please
correct, reject, or discuss as needed. The maintainer is the
decision-maker; nothing here is a requirement.

This lands a draft THREAT_MODEL.md for apache/opendal and wires the
conventional AGENTS.md → SECURITY.md → THREAT_MODEL.md chain so an automated
security scanner can mechanically locate the model.

• THREAT_MODEL.md (new) — a v1 draft modelling OpenDAL's trust boundaries
  (caller→Operator, OpenDAL→backend network, config/credential ingestion,
  FFI), the per-backend credential/auth surface, and the properties OpenDAL
  does / does not provide.
• SECURITY.md (new) — routes vulnerability reports to the ASF security
  process and points at the threat model.
• AGENTS.md — adds a Security section linking SECURITY.md; your existing
  AGENTS.md is otherwise unchanged.

How it was produced. Drafted from OpenDAL's public artifacts (README, the
core/services/ + core/layers/ structure) following the threat-model rubric
at https://gist.github.com/scovetta/2dc9a0695c7cbcc32e23799e00d2ced3. Every
claim is tagged (documented) — from your docs/source — or (inferred) —
our reasoning, not yet confirmed. The inferred claims are collected as
"§14 Open questions for the maintainers"; a one-line confirm/correct per
question is enough, and we fold answers in (inferred → maintainer).

The 15 open questions cover what only the PMC can settle: whether a
malicious/compromised backend is in the adversary model (and how hardened
response/listing parsing is); TLS fallback posture; whether the
logging/tracing layers can ever emit credentials; SSRF / endpoint-trust
responsibility; presigned-URL semantics; the unsafe/FFI memory-safety
posture; and what scanners report that you consider non-findings (§11a).

Context: the ASF Security team is preparing OpenDAL for an automated agentic
security scan being piloted by the team; a discoverable model is the one hard
pre-flight gate. This is the first of the in-scope repos — opendal-reqsign,
opendal-go-services, and opendal-oli follow as separate PRs.

Questions / pushback welcome — happy to adjust file placement or wording to
match the project's house style.

[potiuk]

Closing as a duplicate of #7641, which is the active draft threat-model PR for this repo. Apologies for the noise.