Pool result matches

[sayrer]

This change adds a resultBuf field to nfaBuffers and a matchesInto()
method to matchSet. Instead of allocating a new slice for each match
result, we reuse the pooled buffer by appending matches to it. The
buffer is reset to length 0 before each use while preserving capacity.

See #470.

It only saves one alloc/op, but the bytes/op go way down.

Before:

% go test -bench=^Benchmark8259Example$ -run=^$
FA: Field matchers: 2 (avg size 2.500, max 4)
Value matchers: 5
SmallTables 20371 (splices 6, avg 4.033, max 66, epsilons avg 0.001, max 2) singletons 1
99842/sec
goos: darwin
goarch: arm64
pkg: quamina.net/go/quamina
cpu: Apple M1 Ultra
Benchmark8259Example-20    	  134223	     10016 ns/op	     904 B/op	      23 allocs/op

After:

% go test -bench=^Benchmark8259Example$ -run=^$
FA: Field matchers: 2 (avg size 2.500, max 4)
Value matchers: 5
SmallTables 20371 (splices 6, avg 4.033, max 66, epsilons avg 0.001, max 2) singletons 1
100248/sec
goos: darwin
goarch: arm64
pkg: quamina.net/go/quamina
cpu: Apple M1 Ultra
Benchmark8259Example-20    	  127650	      9975 ns/op	     680 B/op	      22 allocs/op

[timbray]

saves one allocation and 41nsec per op. We're getting into diminishing returns. One question in the review.

[timbray]

Might the builtin copy() work here? Might even be a tiny bit faster.

[timbray]

oh sorry, of course not, m.set is a map not a slice

[timbray]

oh sorry, of course not, m.set is a map not a slice

[sayrer]

It's a big win in memory footprint, even if only modest in CPU:

=== Heap Profile Comparison: main vs PR3 (pool-result-buffer) ===

TOTAL MEMORY ALLOCATED:
Main: 642.27 MB
PR3: 465.26 MB
SAVED: 177.01 MB (-27.6%)

TOTAL ALLOCATION OBJECTS:
Main: 12,775,054 allocations
PR3: 11,116,748 allocations
SAVED: 1,658,306 allocations (-13.0%)

KEY CHANGES:

matchSet.matches() (eliminated by pooling):
Main: 149.03 MB (23.20%), 651,139 allocations (5.10%)
PR3: ELIMINATED (0 MB, 0 allocations)
IMPACT: -149.03 MB, -651,139 allocations

transmap.all:
Main: 115.50 MB (17.98%), 4,325,491 allocations (33.86%)
PR3: 101.50 MB (21.82%), 3,776,613 allocations (33.97%)
SAVED: -14 MB, -548,878 allocations (-12.7%)

traverseNFA:
Main: 348.64 MB cumulative
PR3: 335.16 MB cumulative
SAVED: -13.48 MB (-3.9%)