Cover single subscribe/unsubscribe (by email and by userId), batch bulk
subscribe with per-item multistatus, and the client-side validation paths
(missing identifier, >6 items, differing subscriptions in a batch). Group IDs
are read from env (E2E_ITERABLE_MESSAGE_CHANNEL_ID / _MESSAGE_TYPE_ID /
_EMAIL_LIST_ID) rather than hardcoded; document all required env vars in the
destination's __e2e__/index.ts.

The /api/subscriptions/{group}/{id}/... endpoints are gated per Iterable
project and must be enabled by Iterable, so the HTTP fixtures fail with a 404
"Endpoint not found for project" until that is done. The verboseFailureHint on
those fixtures explains this so the failure is self-diagnosing.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Copies the LinkedIn DMP Company Segment sync action (account-based
marketing) from linkedin-company-audience-hooks (PR #3856) so we can
build e2e-framework tests against it.

- New updateCompanyAudience action + api.ts methods + index.ts registration
- Deletes the destination-level snapshot.test.ts (incompatible with
  hook-gated actions), matching the source PR.
- Intentionally omits metadata.json: this branch deliberately strips
  metadata.json from all destinations (kept only for amplitude).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Collapse duplicate company+action payloads within a batch into a single
element sent to LinkedIn (keyed on normalized domain + organization URN,
so case, whitespace, and bare-id-vs-URN forms of the same company match).
LinkedIn's per-item result is fanned back out to every original payload
index via a parallel index-list, so each input row still gets its own
MultiStatusResponse entry.

Expands the e2e duplicate fixture to interleave non-contiguous dupes,
all three normalization paths, and two invalid rows.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Covers the same scenario as the e2e duplicate fixture: 12 input rows
collapse to 4 unique company+action elements, and each LinkedIn result
(success and error) is fanned back to every original index. Asserts the
exact request body and the full per-item response array, exercising
case/whitespace/bare-id-vs-URN normalization and interleaved no-identifier
validation failures.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Adds a batchWithMultistatus fixture that syncs 350 real company domains
to the DMP Company Segment in a single request (well under LinkedIn's
5000-per-batch max), asserting a 201 for every row.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Adds coverage for the previously-untested hook error/edge paths:
- create segment returns 201 but no x-restli-id -> CREATE_SEGMENT_FAILURE
- create segment request throws -> error result, no savedData
- fetch existing segment throws -> error result, no savedData
- getCompanyAudiences (dropdown) throws -> empty choices + error
- companyAudienceHook.performHook wrapper: delegates to performCompanyHook
  and defaults hookInputs to {} when undefined

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Renames the columns_to_hash field to columns_to_transform and lets each
column select a hash_algorithm ('none' or 'sha256') and a normalize mode
('none' | 'lowercase' | 'trim' | 'lowercase_trim'), both defaulting to
'none'. Normalization runs before hashing via processHashing's cleaning
function, so already-hashed values are never normalized; hash_algorithm
'none' normalizes without hashing. All gated behind the existing
actions-s3-hashing flag.

- Extract SUPPORTED_HASH_ALGORITHMS / SUPPORTED_NORMALIZATIONS to
  constants.ts; generate field choices from them.
- Lock the choice sub-fields to dropdown/typed input only
  (disabledInputMethods).
- Unit tests for every normalize mode, normalize-then-hash, hash with
  normalize none, already-hashed skip, and resolveColumnTransforms
  validation; e2e fixtures for normalize-then-hash and normalize-only.

Verified end-to-end against the real S3 bucket: recomputed every SHA256
and confirmed normalize-only writes transformed plaintext.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Ports the copilot-comment fix from s3-shah256-support: the normalize
field said "Already-hashed values are never normalized", which is
inaccurate for the normalize-only path. Corrected to "Values that are
already hashed are never re-hashed" (the guarantee processHashing
actually provides).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>