fix(photon): preserve text in mixed iMessage attachments (salvage #46513)

[austinpickett]

Summary

Salvage of #46513 (@1960697431 / YuhangLin) onto current main, plus Windows-robustness fixes.

When an iMessage bubble carried both text and an attachment, spectrum-ts' inbound mapper returned only buildAttachmentMessage(...), dropping the user's typed text before Hermes could see it. The Photon adapter also had no group content path, so the text was lost entirely.

Confirmed still present on main: plugins/platforms/photon/adapter.py handles only single text OR attachment/voice content types — no mixed/group flattening — and patch-spectrum-mixed-attachments.mjs doesn't exist.

What changed

• adapter.py: new group content type that flattens text + attachment items, preserving typed text alongside cached media (extracted a shared _normalize_binary_payload helper).
• sidecar: emits group content in normalizeContent, and ships patch-spectrum-mixed-attachments.mjs which patches spectrum-ts' pinned mapper — at npm postinstall AND at sidecar startup, so existing installs self-heal at runtime.
• Tests: test_spectrum_patch.py (patcher regression) + expanded test_inbound.py.

Windows-robustness fixes on top of the original PR

These two bugs made the patcher silently no-op on Windows, so I fixed them as part of the salvage:

1. CLI guard: the patcher used import.meta.url === \file://${process.argv[1]}`, which never matches on Windows (file:///+ drive letter) — invoked as a subprocess it did nothing. Switched topathToFileURL(process.argv[1]).href`.
2. Line endings: the patcher matched \n-joined strings, so a CRLF checkout (Windows git autocrlf) defeated every replacement. It now normalizes CRLF→LF for matching and restores the original EOL style on write.

Without these, test_spectrum_patch.py fails on Windows (the patcher exits 0 but inserts nothing).

Test plan

• pytest tests/plugins/platforms/photon/ → 90 passed. The only failures are 3 pre-existing test_sidecar_lifecycle.py::test_reap_* cases — verified they also fail on pristine main (Windows process-reaping limitation), and this PR doesn't touch that file.
• node --check on the patcher + sidecar → OK
• E2E: ran the patcher against a CRLF fixture — marker + group content inserted, returncode 0, CRLF preserved.

Credit

• fix(photon): preserve text in mixed iMessage attachments #46513 by @1960697431 (YuhangLin) — authorship preserved on the fix commit; Windows fixes + attribution mapping are follow-up commits.

[github-actions]

🔎 Lint report: salvage/photon-mixed-imessage-46513-v2 vs origin/main

ruff

Total: 0 on HEAD, 0 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 0 pre-existing issues carried over.

ty (type checker)

Total: 10990 on HEAD, 10992 on base (✅ -2)

🆕 New issues (1):

Rule	Count
invalid-assignment	1

First entries

tests/run_agent/test_credits_notices_toggle.py:76: [invalid-assignment] invalid-assignment: Object of type `None` is not assignable to attribute `_credits_session_start_micros` of type `int`

✅ Fixed issues (2):

Rule	Count
unresolved-attribute	2

First entries

tests/run_agent/test_credits_notices_toggle.py:76: [unresolved-attribute] unresolved-attribute: Unresolved attribute `_credits_session_start_micros` on type `AIAgent`
run_agent.py:2941: [unresolved-attribute] unresolved-attribute: Object of type `Self@get_credits_spent_micros` has no attribute `_credits_session_start_micros`

Unchanged: 5791 pre-existing issues carried over.

Diagnostics are surfaced as warnings — this check never fails the build.

[OutThisLife]

Approving this salvage. Confirmed the bug is still live on main (the adapter has no group/mixed path, and patch-spectrum-mixed-attachments.mjs doesn't exist), and the fix is sound:

• adapter.py: the new group content type flattens text + attachment items, and the extracted _normalize_binary_payload helper keeps the existing single-attachment/voice behavior byte-for-byte identical. Good refactor.
• The runtime patch is well-guarded: idempotent via the MARKER sentinel; replaceOnce/replaceFirst assert match counts and fail loud on drift; CRLF↔LF normalization handles Windows autocrlf checkouts; and the pathToFileURL(process.argv[1]).href CLI guard fixes the Windows no-op. spectrum-ts being pinned to an exact 3.1.0 is what makes exact-string patching of compiled output acceptable here.
• Attribution preserved (AUTHOR_MAP + co-authored-by); regression tests cover both the patcher and the inbound group path.

One non-blocking consideration: in index.mjs, a patchSpectrumTs() failure at startup process.exit(3)s the whole sidecar. With the exact 3.1.0 pin this won't fire in practice, but if the pin is ever bumped without updating the patch strings, the entire Photon channel goes dark at boot rather than degrading to the (pre-existing) text-dropping behavior for mixed messages only. Failing loud has merit — it surfaces the breakage instead of silently regressing — so I'd leave it as a maintainer judgment call; just flagging the blast radius. Not blocking.