test(drive): remove dead countingBlockFetcher helper

[major0]

Summary

countingBlockFetcher in api/drive/blockstore_internal_test.go is dead code.
It is instantiated in TestBlockStoreCacheMissFetchPopulates and immediately
discarded with _ = fetcher, and its fetch method is never called from
anywhere. The accompanying comment recorded that a real HTTP mock was not
feasible in-package, so the type existed only as commentary.

golangci-lint 2.13.2 reports it via unparam:

api/drive/blockstore_internal_test.go:240:56: (*countingBlockFetcher).fetch - result 1 (error) is always nil (unparam)

That diagnosis is correct rather than a false positive. No interface constrains
the signature, because nothing calls the method, so the error result is provably
always nil. Older golangci-lint releases did not flag it, which is why it went
unnoticed until now.

Changes

Remove the type, its fetch method, the discarded fetcher variable, and the
stale comment referencing it. The io import is retained — it is still used
elsewhere in the file.

Testing

Verified against golangci-lint 2.13.2 and the Go 1.26.8 toolchain, each step
logged in full:

Step	Result
go build ./...	pass
go vet ./...	pass
go test ./...	pass
golangci-lint run --config .golangci.yml	0 issues (was 1)

A full-module lint of main reports exactly this one issue and nothing else, so
this brings the whole module clean.

Scope

Two related things are deliberately not addressed here:

• The test still builds an httpBlockStore it discards the same way
  (_ = store), and its doc comment claims to verify that a cache miss fetches
  over HTTP and populates the disk cache when it only round-trips the buffer
  cache. Making the test match its name is a behavioural change to test
  coverage and belongs in its own PR.
• This is not what unblocks fix(lumo): use the 2.0 chat/completions generation endpoint #12 and feat(lumo): add a --web-search flag to serve #13. Those fail because the pre-commit
  golangci-lint hook compares against HEAD rather than the pull request
  base, so it reports pre-existing findings anywhere in the module regardless of
  what the PR touched. That is a separate CI fix.