feat(parsers+api): 270/271 eligibility request + response (SP3 P4)

Adds API-only 270/271 eligibility flow per spec section 3.4 (no UI, no
DB persistence). Operator pastes a 270 query, gets back raw 270 text,
submits manually, pastes the 271 back.

New models (270 + 271):
  - InformationSource, InformationReceiver, Subscriber, Patient
  - EligibilityBenefitInquiry (EQ) and CoverageBenefit (EB)
  - Inline SERVICE_TYPE_CODES dict (566 codes, snapshot 2026-06-20)
  - service_type_description() lookup helper

Parsers (single segment walker, defensive, mirrors 835 pattern):
  - parse_270: HL*20/21/22/23 loops + EQ, NM1, DMG, TRN
  - parse_271: HL*20/21/22/23 loops + EB, DTP, MSG, AAA

Serializer:
  - serialize_270: ISA/GS/ST/BHT/HL*20/21/22/patient/EQ/SE/GE/IEA
  - round-trips: serialize -> parse -> serialize == identical

API:
  - POST /api/eligibility/request (JSON body -> {raw_270_text, parsed})
  - POST /api/eligibility/parse-271 (file upload -> structured summary)
  - 400 on missing fields / garbage; never 500.

Tests: 365 -> 408 backend (+43 new), 10 frontend unchanged.
TypeScript clean.

backend/tests/fixtures/minimal_270.txt

@@ -0,0 +1,16 @@
+ISA*00*          *00*          *ZZ*SUBMITTERID    *ZZ*RECEIVERID     *240101*1200*^*00501*000000001*0*P*:~
+GS*HC*SUBMITTERID*RECEIVERID*20240101*1200*1*X*005010X279A1~
+ST*270*0001*005010X279A1~
+BHT*0022*13*REF123*20240101*1200~
+HL*1**20*1~
+NM1*PR*2*PAYER NAME*****PI*SKCO0~
+HL*2*1*21*1~
+NM1*1P*2*PROVIDER NAME*****XX*1234567890~
+HL*3*2*22*0~
+TRN*2*TRACE001~
+NM1*IL*1*DOE*JOHN*A***MI*MEMBER123~
+DMG*D8*19800101*M~
+EQ*1^30~
+SE*10*0001~
+GE*1*1~
+IEA*1*000000001~