Count a job's receivers on its own channel, and give text arrays their declared shape

[gdevenyi]

Fixes #201.

Two coupled gaps meant the ACQP-side receiver count of spec 3.3 was never actually computed.

1. Text arrays were delivered flat

GenericParameter.value reshaped numeric arrays to the declared size but returned every text-dtype array flat, so ACQ_ReceiverSelectPerChan declared ( 2, 7 ) arrived as (14,) — with no rows to index.

Spec 2.3 separates the two shapes by element count: a string array carries a final length indicator for the string length (VisuCoreDataUnits=( 2, 65 ) is two strings, not 130), while an enum array carries none (( 2, 7 ) really is two channels of seven receivers). _text_shape picks whichever the count fits. An unrecognised shape stays flat rather than raising — unlike the numeric case, a text array has always been delivered that way. A corpus sweep over 5477 text arrays finds every one matching one rule or the other, none neither.

2. The receiver count ignored chanNum

if channels == fallback:
    return channels
return fallback

made the returned value always PVM_EncNReceivers — and 1 whenever the method omits it — inverting the authority spec 3.3/14.4 give acqp. ACQ_jobs[n][7] was never read.

Now: c = ACQ_jobs[n].chanNum → count Yes in ACQ_ReceiverSelectPerChan row c-1; fall back to the flat ACQ_ReceiverSelect; PVM_EncNReceivers last.

On the six real 2-channel datasets (T2star_map_MGE_mod_all/_mod_pos), ACQ_ReceiverSelectPerChan now arrives as (2, 7), chanNum is read as 1, and the count is 1 — the spec's answer, by the spec's route. All 101 rawdata.* files in the corpus still size exactly.

Tests

• test_a_text_array_takes_its_declared_shape — the enum, string and 1-D cases.
• test_rawdata_receiver_count_follows_the_jobs_channel — a synthetic two-job PV360 experiment where channel 1 has one active receiver and channel 2 has three, and PVM_EncNReceivers disagrees with both. Both fail on master.

Suite: 2142 passed, 12 skipped.

[gdevenyi]

Rebased onto #213 (ruff format), so this branch carries that commit until #213 merges — merge #213 first and this diff reduces to its own change. Verified after the rebase: ruff check and ruff format --check clean, full suite green.

Integration check across all eleven conformance PRs merged together: 2164 passed, 12 skipped; corpus load test 3202 → 3207 of 3478 with zero newly broken datasets.