[ffumero2003]

Context

Address the reader/replication/iterator targets from Part A of #290 on
origin/main (59104c6). Reversing Reader.setErr's first-error guard prevents
cancellation and can strand existing download tests. The other nine specified
mutants are behaviorally equivalent and are skipped below, not suppressed.

Changes

• Add b2/a_reader_error_test.go, with no network, credentials, timers, or worker
  goroutines. The first error must be retained and cancel the context exactly
  once; a later error must neither replace it nor cancel again.
• Put the assertion before transfer tests in unshuffled, fail-fast test order so
  the timeout target reports a prompt assertion failure.
• Change only this reader test file; production and other streams are unchanged.

Target	Disposition	Exact behavior / equivalence reason
reader.go:111, CONDITIONALS_NEGATION	KILLED by the new test	Both the nil-error and existing-error branches are asserted. == nil → != nil fails immediately.
reader.go:266, CONDITIONALS_BOUNDARY	Equivalent; skipped	At available == 0, assigning zero to zero changes nothing.
reader.go:269, CONDITIONALS_BOUNDARY	Equivalent; skipped	At available == r.length, assigning r.length preserves the same value.
reader.go:286, CONDITIONALS_BOUNDARY	Equivalent; skipped	At cr == 1, assigning 1 to 1 preserves the worker count.
reader.go:459, CONDITIONALS_BOUNDARY	Equivalent; skipped	SDK backoff starts at zero, initializes to 1ms, and doubles through 8192ms to 16384ms; the differing 10s equality is unreachable.
replication.go:215, CONDITIONALS_BOUNDARY	Equivalent; skipped	Equal priorities assign the same current maximum.
replication.go:219, CONDITIONALS_BOUNDARY	Equivalent; skipped	At max == maxReplicationPriority - 5, fallthrough adds 5 and returns the same maximum as the cap branch.
replication.go:251, ARITHMETIC_BASE	Equivalent; skipped	len(mapping)+1 → len(mapping)-1 changes only a map allocation hint; all entries are copied unchanged.
iterator.go:102:29, CONDITIONALS_BOUNDARY	Equivalent; skipped	The surviving > 1000 → >= 1000 comparison assigns 1000 to an existing 1000. The separate < 0 comparison is not this survivor.
iterator.go:108, CONDITIONALS_BOUNDARY	Equivalent; skipped	At page size 100, assigning 100 to 100 changes nothing.

No tests manufacture unreachable backoff values or assert map allocation details
to claim kills for the equivalent targets. No Part B coverage changes.

Verification

All commands used empty B2_APPLICATION_KEY_ID, B2_APPLICATION_KEY,
B2_ACCOUNT_ID, and B2_ACCOUNT_KEY.

Hand-applied the exact reader.go:111 mutation alone, then restored it. The first
branch red proof from the new test:

=== RUN   TestReaderSetErrKeepsFirstErrorAndCancelsOnce/first_error
recorded error = <nil>, want first download failure
reader context error = <nil>, want context canceled
cancel calls = 0, want 1

The existing-error branch rejects that same mutation too:

=== RUN   TestReaderSetErrKeepsFirstErrorAndCancelsOnce/existing_error
recorded error = later download failure, want first download failure
cancel calls = 1, want 0
FAIL github.com/Backblaze/blazer/b2 0.511s

The mutated full package also fails promptly with
go test -count=1 -failfast -timeout=20s ./b2:

--- FAIL: TestReaderSetErrKeepsFirstErrorAndCancelsOnce (0.00s)
    --- FAIL: TestReaderSetErrKeepsFirstErrorAndCancelsOnce/first_error (0.00s)
FAIL github.com/Backblaze/blazer/b2 0.313s

After restoring the production guard, the focused test passes:

--- PASS: TestReaderSetErrKeepsFirstErrorAndCancelsOnce (0.00s)
    --- PASS: TestReaderSetErrKeepsFirstErrorAndCancelsOnce/first_error (0.00s)
    --- PASS: TestReaderSetErrKeepsFirstErrorAndCancelsOnce/existing_error (0.00s)
PASS
ok github.com/Backblaze/blazer/b2 0.324s

• go build ./...: PASS, no output.
• go vet ./...: PASS, no output.
• go test ./b2/...: PASS (b2 11.482s; b2/sync 2.761s).
• gofmt -l .: clean.
• make ci: PASS, including lint (0 issues), race/shuffle tests, module
  checks, b2keys, cross-compiles, changelog check, and script tests. The b2 race
  suite passed in 20.021s.
• Fresh review of the final diff and mutation claims: no blocking findings.
• No Gremlins rerun; the proof is the exact hand-applied mutation and restored
  green run. Equivalent mutants were reviewed and skipped, not reported killed.
• Live B2 and simulator checks: NOT RUN. This tests reader error bookkeeping and
  cancellation without changing production request behavior or harness scenarios.

Actual make ci tail (exit 0):

ok   skips finished tests in a failed package
ok   keeps the error line after long failure output
ok   keeps the last lines of long failure output
ok   says how many lines were omitted
ok   keeps the timeout reason above a long stack trace
ok   keeps a timeout reason that follows 150 progress lines
ok   keeps the failure of the first of three failing tests
ok   keeps the failure of the quiet test between two noisy ones
ok   keeps the failure of the last of three failing tests

Part of #290