fix(flights): match aircraft types exactly, not as substrings (609 -> 27 false 'military' contacts)

[ihubanov]

isMilitaryType() matches type designators as substrings:

return MILITARY_AIRCRAFT_TYPES.some(mt => upper.includes(mt));

The list contains short codes (C17, C5, C2, E3, B2), so any ICAO type that merely contains one of them matches. A lot of ordinary civil traffic ends up counted as military.

I pulled a live regional feed for the Iran/Israel theater (lat/30/lon/48/dist/2500, 2754 aircraft) and compared what the heuristics flag against adsb.lol's own dbFlags military bit:

  21   carry the dbFlags military bit
 588   added by the heuristics
~609   reported by the panel

What the substring rule was actually catching:

490x  B738        ordinary 737-800s
 17x  A332
 12x  C172   ->   matched C17, rendered "Strategic Airlift (C-17)"
 11x  C56X   ->   matched C5,  rendered "Strategic Airlift (C-5)"
  6x  E35L   ->   matched E3,  rendered "AWACS"
  8x  C25A/B ->   matched C2
  3x  PC24   ->   matched C2

So Cessna 172s were showing up as C-17 Globemasters and Embraer Legacy bizjets as E-3 AWACS. classifyAircraft() has the same problem; its t.includes('B2') branch labels any Beechcraft King Air (B200) a bomber.

isMilitaryCallsign() matches bare prefixes, so SAMU42 and SAMU06 (French SAMU air ambulances) match the US SAM Special Air Mission prefix.

Changes

MILITARY_AIRCRAFT_TYPES is a Set now, matched exactly. Converting it turned up several entries that aren't real ICAO Doc 8643 designators and so never matched anything: KC135 should be K35R, RC135 is R135, CH47 is H47, AH64 is H64, MQ9 is Q9, RQ4 is Q4. The list was over-matching civil types and missing real military ones at the same time.

Dropped the dual-use codes: B737/B738/B739, A332, A310, GLF5, GLF6, A124, C40. Military examples of those airframes already carry the dbFlags bit, so the heuristic wasn't buying anything. Concretely: PLF110, a Polish Air Force 737, is still picked up after removing B737 from the list. That's the case the entry was there for.

classifyAircraft()'s type chain is an exact TYPE_ROLES lookup instead of the includes() ladder.

isMilitaryCallsign() now requires a digit after the prefix. RCH471 and PLF110 still match, SAMU42 doesn't. I also dropped the bare 5-6 digit rule, which was a guess and didn't match anything in the sample.

Removed the A00000-AFFFFF + type branch. The unconditional isMilitaryType() check above it already returns for anything that branch could catch, so it was dead.

Result

609 -> 27 on the same feed, 25 of them dbFlags-confirmed. What's left is Spanish A400Ms, a German A321neo, Italian and Belgian air force aircraft, US Army CH-47s and a UH-72, a KC-135R (57-1468), French Sécurité Civile Dash-8s, a Croatian CL-415. No Cessnas or business jets.

No API or response-shape changes, only the classification predicates. next build passes.

One thing I left alone: several entries in MILITARY_CALLSIGN_PREFIXES are broad enough to collide with civil callsigns even with the digit boundary (MAG, KING, GOLD, CLUB, TANGO). Probably worth a separate look.