perf: skip lid blocks

[cheb0]

Description

Summary

• new frac version: v4
• LID block ext1 now has firstLID and lastLID encoded
• isContinued is only used for older frac versions
• no legacy format TID adjusting for a new frac version
• as previously, we always locate and load the first block eagerly but calling NextGeq allows to skip some blocks

Measurements

With 16k LID blocks, there are only 3 search requests (out of 50 I tested) where this feature improves cold query perf. All three search requests fit into "intersect dense and sparse column" type.

With 4k blocks there are more queries that speed up (even some aggs), but it's not sufficient to cover perf slowdown from using 4k LID block instead of 64k.

Query	Type	Env	Ids	Total	cold, ms	stddev	cold (branch), ms	stddev	diff	Trial 1	Trial 2	Trial 3	Trial 4
service:E and login:L and moduleName:S and (zone:A or zone:B or zone:C) and exists:message	srch	prod	8	0	17.04	±1.64	13.91	±0.54	-18.4%	-18.5%	-17.6%	-19.9%	-18.4%
request_host:huge AND remote_addr:X (found)	srch	lb	5	0	12.43	±0.43	11.35	±0.51	-8.7%	-8.1%	-5.4%	-5.8%	-8.7%
request_host:small AND response_status:200	hist	lb	69	69	25.58	±1.04	20.9	±0.60	-18.3%	-17%	-17.9%	-14.1%	-18.3%

---

• I have read and followed all requirements in CONTRIBUTING.md;
• I used LLM/AI assistance to make this pull request;