fix(terser): share nameCache between chunks by serializing minify tasks

[itsjamie]

Rollup Plugin Name: terser

This PR contains:

• bugfix
• feature
• refactor
• documentation
• other

Are tests included?

• yes (bugfixes and features will not be merged without tests)

Breaking Changes?

• no

List any relevant issue numbers:

resolves #1970

Description

The plugin already returns the mutated nameCache from its workers and merges it back, but the merge never takes effect across chunks. Rollup calls renderChunk for every chunk concurrently, so each chunk's options are serialized with the still-empty cache before any other chunk's cache comes back. maxWorkers: 1 doesn't help because the pool dispatches the next queued task synchronously from the worker's message handler, before the awaiting merge-back microtask runs. The result is the divergence in #1970: the same method mangled to .o() in one chunk and .m() in another.

Sharing a name cache only works if each minify() call sees the names the previous call assigned, so when nameCache is provided the minify tasks now run through a promise chain: each chunk's options are snapshotted and serialized only after the previous chunk's cache has been merged back. Builds without a nameCache keep the current parallel behaviour.

Tests cover a shared cache across two entries importing a common module (the shape from the issue) and across two unrelated entries, each with default workers and with maxWorkers: 1. Two more tests pin the surrounding behaviour: without a nameCache unrelated chunks still mangle independently, and with one the build uses a single worker even when maxWorkers: 4 is set.

A side effect worth a second look: a shared nameCache now effectively serializes the whole build onto one worker. I documented that in the README rather than emitting a runtime warning. Open to thoughts on whether it deserves more than a docs note.