Drop synthetic-job / event-stream-bridge framing for 7a-2+

[bdraco]

What does this implement/fix?

Drops the "synthetic offloader-side FirmwareJob + event-stream
bridge" framing from docs/ARCHITECTURE.md's description of
the transparent install flow's 7a-2 / 7a-3 / 7a-4 slices.
That shape was inherited from the original design doc and
reverse-engineered from "what's the minimum change to avoid
touching the install dialog's subscriber?" — a bad starting
point. It created two FirmwareJobs per remote build (one on
each side), a bridge that re-fired OFFLOADER_JOB_* events as
JOB_* against a synthetic id, and a separate RAM-only id
mapping. Extra ceremony for a problem that doesn't exist —
lockstep frontend deployment means the install dialog can
learn the new field (job.source_label) in the same
coordinated PR that lands the source-routed runner.

What's replacing it

One FirmwareJob per build, the runner branches on
source to pick its pipeline (local subprocess vs peer-link
submit_job dispatch). Same JOB_* lifecycle events fire
either way; the install dialog stays on its existing
subscriber. OFFLOADER_JOB_* events remain as the wire-layer
fan-out (5c-3 keeps them for the explicit Send-builds dialog
that hangs around as the power-user surface); the runner
consumes them privately to update its single FirmwareJob.

Renumbered slices to reflect the new shape:

• 7a-2a: FirmwareJob.source / source_pin_sha256 /
  source_label field additions. In flight as Add FirmwareJob.source / source_label fields (7a-2a) #556.
• 7a-2b: source-routed runner (renamed from
  "event-stream bridge"; the bridge concept disappears
  entirely). The firmware queue's _run_job gains a branch on
  job.source; LOCAL keeps the existing esphome run
  subprocess; REMOTE calls PeerLinkClient.submit_job and
  subscribes to OFFLOADER_JOB_* filtered to its dispatch's
  job_id, updating the same FirmwareJob's status as wire
  events arrive.
• 7a-3: firmware/install integration. Unchanged in
  spirit but now constructs a regular FirmwareJob with
  source=REMOTE rather than a synthetic wrapper.
• 7a-4: cancel translation. Simplifies — the runner's
  cancel hook fires remote_build/cancel_job directly. No
  id translation between two FirmwareJobs because there
  aren't two.

Added a leading paragraph above the slices that explains why
the synthetic framing was wrong, so future readers don't have
to reconstruct the reasoning from PR history.

Related coordinated changes

• Add FirmwareJob.source / source_label fields (7a-2a) #556 (7a-2a) is the load-bearing model change for the
  first-class shape. This PR's arch.md description references
  the fields it adds.
• Issue Remote build offload: discover and delegate compiles to another dashboard on the LAN #106 § Transparent install flow is the source of
  truth for the full design; a follow-up edit will bring its
  body into alignment with this PR's framing (no PR needed —
  just gh issue edit).

Related issue or feature:

• Documentation catch-up for Remote build offload: discover and delegate compiles to another dashboard on the LAN #106
  (transparent install flow). Companion to model PR Add FirmwareJob.source / source_label fields (7a-2a) #556.

Types of changes

• Bugfix (non-breaking change which fixes an issue) — bugfix
• New feature (non-breaking change which adds functionality) — new-feature
• Enhancement to an existing feature — enhancement
• Breaking change (fix or feature that would cause existing functionality to not work as expected) — breaking-change
• Refactor (no behaviour change) — refactor
• Documentation only — docs
• Maintenance / chore — maintenance
• CI / workflow change — ci
• Dependencies bump — dependencies

Frontend coordination

• No frontend change needed
• Companion frontend PR: esphome/device-builder-dashboard-frontend#

Checklist

• The code change is tested and works locally.
• Pre-commit hooks pass (ruff, codespell, yaml/json/python checks).
• Tests have been added or updated under tests/ where applicable — N/A; docs-only.
• components.json has not been hand-edited.
• Architecture-level changes are reflected in docs/ARCHITECTURE.md and/or docs/API.md — this PR is the architecture-level reflection.

[codspeed-hq]

Merging this PR will not alter performance

✅ 12 untouched benchmarks

---

_{Comparing docs/arch-drop-synthetic-state (f58fe49) with main (ca1ce69)}

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.12%. Comparing base (ffb8274) to head (e45b810).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #558   +/-   ##
=======================================
  Coverage   99.12%   99.12%           
=======================================
  Files          77       77           
  Lines       10033    10033           
=======================================
  Hits         9945     9945           
  Misses         88       88           

Flag	Coverage Δ
py3.12	99.09% <ø> (ø)
py3.14	99.12% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull request overview

Updates the transparent install flow architecture documentation to remove the “synthetic FirmwareJob + event-stream bridge” framing and describe the newer “one FirmwareJob per build, source-routed runner” model for slices 7a-2+.

Changes:

• Adds rationale explaining why the synthetic-job / event-translation bridge approach was unnecessary.
• Renumbers and rewrites the in-flight slices as 7a-2a / 7a-2b / 7a-3 / 7a-4 to match the first-class FirmwareJob.source shape.
• Updates the described event flow so OFFLOADER_JOB_* stays as wire fan-out, while the runner privately consumes it and continues emitting JOB_* for the install dialog.