fix(trek): integer decode for quadrant K/B/S counts (fixed-point hang)

EnterQuadrant decoded the packed quadrant code with float tricks:
  K3=int(G*.01) : B3=int(G*.1)-10*K3 : S3=G-100*K3-10*B3

Fine under the float interpreter, fatal when transpiled to cc65 fixed-point:
16.16 RGC_MUL mangles .01/.1 (keeps ~2 fractional bits) and G/100 overflows
the 32-bit fixed range for G>=256, so K3 undercounts and S3 = G-100*K3-...
OVERcounts. FindEmpty() (retry-until-empty-sector) then tries to place more
stars than the 64-sector quadrant holds and spins forever — trek hung right
after the mission briefing on a real C64.

Decode with integer \ and MOD instead (G encodes K*100+B*10+S):
  K3=G\100 : B3=(G\10) MOD 10 : S3=G MOD 10
Exact, no overflow, cheap — and behaviour-identical in the interpreter.
Lesson: integer logic wants integer ops, not float arithmetic, on 8-bit.

Author: omiq

examples/trek-portable.bas

@@ -203,10 +203,14 @@ function EnterQuadrant()
         print "ENTERING ";QN$;" QUADRANT..."
       end if
     end if
-    ' Decode G(quad): K3/B3/S3 = klingons / starbases / stars in this quadrant
-    K3=int(G(QUADRANT_X,QUADRANT_Y)*.01)
-    B3=int(G(QUADRANT_X,QUADRANT_Y)*.1)-10*K3
-    S3=G(QUADRANT_X,QUADRANT_Y)-100*K3-10*B3
+    ' Decode G(quad): K3/B3/S3 = klingons / starbases / stars in this quadrant.
+    ' G encodes the 3 digits K*100+B*10+S. Use INTEGER divide/MOD, not float
+    ' tricks (int(G*.01)): on fixed-point targets (cc65) G*.01 loses precision
+    ' and G/100 overflows 16.16, mis-decoding the counts -> FindEmpty overfills
+    ' the quadrant and spins forever. Integer ops are exact and cheap.
+    K3=G(QUADRANT_X,QUADRANT_Y)\100
+    B3=(G(QUADRANT_X,QUADRANT_Y)\10) MOD 10
+    S3=G(QUADRANT_X,QUADRANT_Y) MOD 10
     for I=1 to 3
       K(I,1)=0
       K(I,2)=0