core/rawdb: implement sequential reads in freezer_table

[holiman]

This PR is a bit of a successor to #20308, which implemented sequential access to freezer data via an iterator interface.
This PR instead returns a slice of data. It does not defer reads until Next(), so instead does all the reads while holding the readlock.
It also switches the implementation of Retrieve to be a just a short-cut for doing a sequential read with max items set to 1. This PR doesn't really add any uses of the new method (except the Retrieve), but does actually improve one thing as-is: previously, two syscalls were performed to read two index items from the index-file. This PR instead reads all (both) index items using only one syscall.

Later on, this can be used to serve eth64 data such as headers directly from ancients with a minimum number of syscalls. If that is done, we can get by with 2 syscalls per "192-header-delivery" instead of 576.

[holiman]

I made a little benchmark tool for this, in this branch.

Using the old reads, where each header is read separately, and causing 3 syscalls per header (so 576 syscalls per 192 headers):

192 headers in 962.083µs: 199.57 headers/ms
688128 headers in 5.000379735s: 137.62 headers/ms
1617984 headers in 10.00078688s: 161.79 headers/ms
2851968 headers in 14.999695984s: 190.14 headers/ms
4238400 headers in 19.998605178s: 211.93 headers/ms
5996928 headers in 24.997846647s: 239.90 headers/ms
8137536 headers in 29.996676305s: 271.28 headers/ms
10647360 headers in 34.995323103s: 304.25 headers/ms
13380480 headers in 39.994063183s: 334.56 headers/ms
16335360 headers in 44.993105166s: 363.06 headers/ms
19403136 headers in 49.991620151s: 388.13 headers/ms
22560576 headers in 54.990467372s: 410.26 headers/ms
25768320 headers in 59.988958187s: 429.55 headers/ms
28992768 headers in 1m4.987618054s: 446.13 headers/ms
32201472 headers in 1m9.986038652s: 460.11 headers/ms
35439360 headers in 1m14.984864442s: 472.62 headers/ms
38681856 headers in 1m19.983580689s: 483.62 headers/ms
41926272 headers in 1m24.98204837s: 493.35 headers/ms
45177216 headers in 1m29.980422998s: 502.08 headers/ms
48415680 headers in 1m34.978883429s: 509.75 headers/ms
51676800 headers in 1m39.97733937s: 516.89 headers/ms
54936768 headers in 1m44.975900812s: 523.33 headers/ms
58201344 headers in 1m49.974443941s: 529.23 headers/ms
61437888 headers in 1m54.973013666s: 534.37 headers/ms
64694400 headers in 1m59.971581851s: 539.25 headers/ms
67878528 headers in 2m4.970055655s: 543.16 headers/ms
71136576 headers in 2m9.968586182s: 547.34 headers/ms
74400192 headers in 2m14.96715173s: 551.25 headers/ms
77652096 headers in 2m19.965662809s: 554.79 headers/ms
80911488 headers in 2m24.964044114s: 558.15 headers/ms
84173184 headers in 2m29.962889512s: 561.29 headers/ms
87436416 headers in 2m34.961447644s: 564.25 headers/ms
90681792 headers in 2m39.959987835s: 566.90 headers/ms
93945792 headers in 2m44.958553317s: 569.51 headers/ms
97208064 headers in 2m49.957255953s: 571.96 headers/ms
100463424 headers in 2m54.955884838s: 574.22 headers/ms
103725120 headers in 2m59.954570697s: 576.40 headers/ms
106978176 headers in 3m4.953132496s: 578.41 headers/ms
110243712 headers in 3m9.951801007s: 580.38 headers/ms
113503296 headers in 3m14.950283968s: 582.22 headers/ms
116766528 headers in 3m19.948830709s: 583.98 headers/ms
120026304 headers in 3m24.947243624s: 585.64 headers/ms

It speeds up continously (it's on a NUC with pretty hefty RAM, so the disk reads will be better cached) but after three and a half minute, it's reached 585 headers per millisecond.

If the sequential-read mechanism in this PR is used instead, which does 2 (instead of 576) syscalls per 192 headers , we get

./build/bin/geth db bench2 10000000
INFO [06-29|14:28:05.085] Maximum peer count                       ETH=50 LES=0 total=50
INFO [06-29|14:28:05.085] Smartcard socket not found, disabling    err="stat /run/pcscd/pcscd.comm: no such file or directory"
INFO [06-29|14:28:05.085] Set global gas cap                       cap=50,000,000
INFO [06-29|14:28:05.085] Opening freezer                          location=/home/martin/.ethereum/geth/chaindata/ancient name=headers
192 headers in 125.972µs: 1524.15 headers/ms
9017856 headers in 4.994948262s: 1805.40 headers/ms
18025920 headers in 9.990446257s: 1804.32 headers/ms
27043200 headers in 14.98572444s: 1804.60 headers/ms
36030528 headers in 19.980900692s: 1803.25 headers/ms
45035328 headers in 24.976188053s: 1803.13 headers/ms
54063360 headers in 29.97161218s: 1803.82 headers/ms
63072576 headers in 34.967231597s: 1803.76 headers/ms
72077952 headers in 39.962521288s: 1803.64 headers/ms
81045120 headers in 44.957913329s: 1802.69 headers/ms
90033600 headers in 49.953242546s: 1802.36 headers/ms
99019392 headers in 54.948543408s: 1802.04 headers/ms
108010944 headers in 59.943880611s: 1801.87 headers/ms

[karalabe]

Mostly nitpicks, but there's one corner case that I believe is broken.

[karalabe]

SGTM