perf: batch opencode grep message reads

[sorafujitani]

Summary

• batch OpenCode grep body checks for candidates whose LIKE prefilter is disabled
• preserve projection-over-parts semantics while avoiding one Messages(id, 0) call per candidate
• bound Messages(id, limit) SQL reads for projection and parts stores
• keep total/start time for preview headers via separate count/start queries
• add regression tests for disabled-prefilter batch grep, projection priority, and limit boundary behavior

Fixes #103

Benchmark

Commands:

go test -run '^$' -bench='Benchmark(GrepKeysHitAndMiss|GrepKeysPrefilterDisabled|MessagesLargeSession)' -benchmem ./internal/opencode -count=10

Disabled-prefilter grep baseline was measured on origin/main with only BenchmarkGrepKeysPrefilterDisabled added temporarily.

GrepKeysPrefilterDisabled/hit-12   184.707m ± 2%   2.771m ± 1%  -98.50% (p=0.000 n=10)
GrepKeysPrefilterDisabled/miss-12  183.459m ± 2%   2.783m ± 1%  -98.48% (p=0.000 n=10)

GrepKeysPrefilterDisabled/hit-12   1764.0Ki ± 0%  819.0Ki ± 0%  -53.57% (p=0.000 n=10)
GrepKeysPrefilterDisabled/miss-12  1763.8Ki ± 0%  818.8Ki ± 0%  -53.58% (p=0.000 n=10)

GrepKeysPrefilterDisabled/hit-12   42.54k ± 0%  22.73k ± 0%  -46.57% (p=0.000 n=10)
GrepKeysPrefilterDisabled/miss-12  42.54k ± 0%  22.73k ± 0%  -46.57% (p=0.000 n=10)

Existing benchmark comparison:

GrepKeysHitAndMiss/hit-12   26.50m ± 1%  26.21m ± 1%  -1.12% (p=0.005 n=10)
GrepKeysHitAndMiss/miss-12  27.24m ± 1%  26.88m ± 0%  -1.31% (p=0.000 n=10)
MessagesLargeSession-12      2.979m ± 1%  4.498m ± 3%  +50.96% (p=0.000 n=10)

MessagesLargeSession-12  1106.52Ki ± 0%  36.95Ki ± 0%  -96.66% (p=0.000 n=10)
MessagesLargeSession-12  30057.0 ± 0%     977.0 ± 0%    -96.75% (p=0.000 n=10)

MessagesLargeSession uses small synthetic bodies, so the extra count/start query costs more wall time than it saves in this microbenchmark. The bounded path still avoids fetching all message rows for preview and cuts memory/allocations by about 97%, which is the intended behavior for large real sessions.

Verification

go test ./...
go vet ./...
golangci-lint run
gofmt -l $(git ls-files '*.go')
git diff --check