Return the unique ID from single-message UIDL replies

[OskarEichler]

Summary

Return the unique ID, rather than the message number, from a single-message UIDL reply. A reply is +OK message-number unique-id; the current parser returns the second token. Use whitespace splitting and take the third token, retaining the ID's case and bytes.

Reproduction

require 'net/pop'
socket = Object.new
socket.instance_variable_set(:@responses, ['+OK fixture', '+OK 2 QhdPYR:00WBw1Ph7x7'])
def socket.readline = @responses.shift
def socket.writeline(*) = nil
command = Net::POP3Command.new(socket)
p command.uidl(2)
# Before: "2"
# After:  "QhdPYR:00WBw1Ph7x7"

This follows the single-message UIDL example in RFC 1939 section 7.

Verification

• 148 focused checks, including a bounded real localhost session: message numbers, punctuation, mixed case, one-/70-character IDs, whitespace, transmitted commands, negative replies and session cleanup. Before: 97 failed expectations; after: zero.
• 200 additional checks retain multiline UIDL/LIST behavior across every split point of the sample replies, including CR/LF boundaries and byte-at-a-time input. These pass before and after with net-protocol 0.3.0 and the consumer's existing protocol fixes.
• Existing suite on baseline and isolated patch: 10 tests / 17 assertions / zero failures/errors, Ruby 4.0.6 via rbenv. Syntax and git diff --check pass; supplemental lint was compared with baseline.
• No repository tests, dependencies, workflow settings or unrelated source changed. Reproduction/check scripts are external under the consumer repository's no-new-tests policy.

Compatibility and limits

No API signature change. The internal POP3Command#uidl(number) intentionally returns the correct ID instead of a message number; callers relying on the wrong value will observe a change. The multiline path used by POPMail#unique_id is unchanged. No new strict malformed-response validation is introduced.

This is separate from issue #27's reported multiline chunk loss. That issue did not reproduce in the bounded current-reader checks above; this PR does not claim to resolve it. Other Ruby/OS runtimes and external mail providers were not tested; no production or external mailbox access occurred.