perf: wrap SQL queries in transaction

[onmax]

Fixes #3669

Problem

During development, processCollectionItems executes SQL INSERT queries one at a time in a loop. SQLite by default wraps each statement in an implicit transaction with fsync to disk. With 300+ queries, this causes ~3.8 seconds of overhead.

Solution

Wrap the SQL execution loop in an explicit transaction (BEGIN TRANSACTION / COMMIT). This batches all queries into a single transaction with one fsync at the end.

Important: D1 doesn't support transactions, so we conditionally skip the transaction wrapper for D1 databases by adding a supportsTransactions flag to the database interface.

Benchmark

Metric	Before	After
SQL exec (323 queries)	3864ms	51ms
Content processing total	4681ms	~1000ms

StackBlitz

	Link	Expected
Bug	content-3669	❌ Slow processing (~3-4s)
Fix	content-3669-fixed	✅ Fast processing (~200ms)

CLI Reproduction

git clone --depth 1 --filter=blob:none --sparse https://github.com/onmax/repros.git
cd repros && git sparse-checkout set content-3669
cd content-3669 && pnpm i && pnpm dev

Verify fix

git sparse-checkout add content-3669-fixed
cd ../content-3669-fixed && pnpm i && pnpm dev

The -fixed folder uses pnpm patch to apply the fix locally.

Changes

• Added supportsTransactions property to LocalDevelopmentDatabase interface
• Set supportsTransactions: database.type === 'sqlite' in getLocalDatabase
• Wrap SQL execution in transaction when supportsTransactions is true

Why this works

SQLite by default wraps each statement in an implicit transaction. When you run 323 INSERT statements individually, SQLite does 323 separate transactions (with fsync to disk each time). Wrapping them in a single explicit transaction does one fsync at the end.

Discovered in: https://github.com/nimiq/nimiq-website/tree/nuxt-content

[vercel]

@onmax is attempting to deploy a commit to the Nuxt Team on Vercel.

A member of the Team first needs to authorize it.

[pkg-pr-new]

npm i https://pkg.pr.new/@nuxt/content@3670

commit: 7d714d6

[farnabaz]

Thanks 👍