Add cps_check_codes() warning and document 2022+ No Response discontinuity

[Franky58-art]

Thanks for reporting this, Dr. Hanmer — confirmed, and it turns out to predate 2024.

The -9 ("No Response") code is entirely absent from PES1/VRS_VOTE in both the 2022 and 2024 data, not just 2024. Reading the full raw files and tabulating every code present in PES1:

Year	-9 (No Resp)	-1 (Not in Univ)	1 (Yes)	2 (No)
2016	6.6%	38.3%	38.5%	14.6%
2018	8.1%	38.0%	33.8%	17.9%
2020	7.6%	38.9%	41.1%	10.5%
2022	0%	48.9%	31.2%	17.8%
2024	0%	48.5%	38.2%	11.6%

The adult share of -1 roughly doubles right when -9 disappears. This suggests the Census Bureau folded "No Response" adults into the generic "Not in Universe" bucket starting in 2022, without documenting the change.

Age composition of the -1 group:

Year	-1 count	% of -1 group aged 18+
2016	58,302	13.1%
2018	54,301	12.6%
2020	52,224	11.6%
2022	61,611	23.4%
2024	61,494	22.4%

Why this matters for cpsvote: cps_label() treats -1 as missing (NA) before cps_recode_vote() ever runs. Through 2020, a -9 respondent was counted as "No" for cps_turnout and correctly dropped to NA for hurachen_turnout. Since 2022, that same person (now coded -1) is wiped to NA at the labelling step and never reaches the recode logic. cps_turnout's denominator silently shrinks, inflating the naive estimate.

Numbers for 2024:

• Naive turnout, PWSSWGT-weighted, excluding only -1s: 73.4% — matches the ~74% figure in the issue
• Hur-Achen adjusted turnout: 62.8% — in the neighborhood of the ~65% CPS-reported rate

Changes made in this PR:

• Added cps_check_codes(), which warns when a documented code is completely absent from the data. cps_recode_vote() now calls this automatically.
• Documented the discontinuity in vignette("voting").

Not yet addressed: using respondent age to split the -1 bucket back into "Not in Universe" vs. "No Response" for 2022+. That would be a substantive methodological change held for discussion. Input welcome.