Add threat model + security-model discoverability (AGENTS.md → SECURITY.md → THREAT_MODEL.md)

[potiuk]

Add the Apache bRPC threat model + security-model discoverability

This lands the bRPC threat model in-repo and wires up the discoverability
chain so automated security tooling (and human triagers) can mechanically find
it:

• THREAT_MODEL.md — the threat model, first drafted by the ASF Security
  Team and revised by @wwbmmm (Weibing Wang) on the PMC side. This PR puts
  that reviewed model under version control in the repo rather than in a gist.
• SECURITY.md — points reporters at the threat model + the ASF security
  reporting process.
• AGENTS.md — a Security section linking SECURITY.md → THREAT_MODEL.md
  so an automated reviewer can follow the chain.

This is a draft for the PMC to finalize — review welcome here

@wwbmmm already did a substantial revision pass (streamlined the original
draft, answered the §14 open questions). Landing it in-repo as a PR is the
"one more look" before it's final — please refine directly here.

Two simplifications @wwbmmm flagged are best done as review on this PR (left to
the author rather than applied unilaterally):

• Collapse the remaining redundancy — e.g. the -immutable_flags /
  /flags point currently appears in the §5a config table, the §5a
  insecure-default summary, §9 D13, §11 M14, and §14 Q46. Decide which homes
  are load-bearing and trim the rest.
• Keep it at the trust-boundary level. Favour high-level statements of how
  trust boundaries are defined and what security properties bRPC does / does
  not provide, over code-level detail that drifts across bRPC versions — so the
  document stays accurate and maintainable as the code evolves. (The §5a tables
  already carry "example, not a complete list — analyze the latest code"
  disclaimers in that spirit.)

The model is version-bound (based on master around 2026-05-21); reports
should be triaged against the model for the corresponding release, and the
document re-bound on each release.

This is part of an automated agentic security-scan pilot the ASF Security team
is running; a discoverable, in-repo threat model lets the scan focus on real
issues and suppress the by-design ones. No project source is touched.

[potiuk]

This is the right model proposal :)

[potiuk]

Reopened — this is the canonical bRPC threat-model PR (it carries @wwbmmm's revised model). #3322 was the duplicate and is now closed; please disregard my earlier "duplicate of #3322" note above and keep the review here.