Make discriminated unions significantly faster

[vbudovski]

There needs to be some special handling for validating discriminated unions specifically, as the benchmarks below show. Regular unions do not exhibit this drastic drop on performance when matching a second or subsequent member.

benchmark                time (avg)        iter/s             (min … max)       p75       p99      p995
------------------------------------------------------------------------- -----------------------------

group Discriminated union 1
Paseri                   55.36 ns/iter  18,064,676.7   (48.2 ns … 129.57 ns) 55.74 ns 87.17 ns 98.02 ns
Zod (regular)           403.06 ns/iter   2,481,003.8   (341.05 ns … 1.18 µs) 408.11 ns 635.94 ns 1.18 µs
Zod (discriminator)     401.87 ns/iter   2,488,366.0 (347.92 ns … 645.58 ns) 409.87 ns 608.34 ns 645.58 ns
Valita                   59.23 ns/iter  16,882,259.3  (51.65 ns … 149.89 ns) 59.24 ns 90.23 ns 97.28 ns

summary
  Paseri
   1.07x faster than Valita
   7.26x faster than Zod (discriminator)
   7.28x faster than Zod (regular)

group Discriminated union 2
Paseri                  257.47 ns/iter   3,883,929.6  (224.5 ns … 406.37 ns) 265.96 ns 339.28 ns 364.05 ns
Zod (regular)             1.89 µs/iter     530,222.7     (1.46 µs … 3.02 ms) 1.75 µs 5.17 µs 9.43 µs
Zod (discriminator)     494.33 ns/iter   2,022,946.7  (441.49 ns … 695.7 ns) 503.44 ns 640.07 ns 695.7 ns
Valita                   76.52 ns/iter  13,067,883.5  (68.04 ns … 122.91 ns) 77.33 ns 108.37 ns 115.75 ns

summary
  Valita
   3.36x faster than Paseri
   6.46x faster than Zod (discriminator)
   24.65x faster than Zod (regular)