left aligned, adaptive, rolling max

[jangorecki]

This PR is fresh attempt of #5441 which has been split in multiple smaller PRs.
10 smaller PRs have been already individually reviewed and merged to this dev branch, which now here is proposed to master.

This PR will supersede the following PRs: #5441, #5890, #5891, #5892, #5893, #5894, #5895, #5896, #5897, #5898

Scope

• adaptive && align="left" Adaptive, left-aligned rolling max #5438
• frollmax adaptive (exact) Adaptive, left-aligned rolling max #5438
• frollmax (fast, exact) rolling functions, rolling aggregates, sliding window, moving average #2778
• give.names argument rolling functions, rolling aggregates, sliding window, moving average #2778
• frollapply adaptive Adaptive, left-aligned rolling max #5438
• de-duplicate logic (by adding helper function - partial2adaptive, coerceK, ansSetMsg, or redirecting frollapply to froll) of
  • align
  • early stopping for window bigger than x
  • partial argument support
  • left adaptive support
  • warnings, errors and some messages
  • k arg handling
• algo="fast" support for Inf and -Inf (breaking change).
• rename hasNA to has.nf (has non-finite)
• partial argument frollmean partial argument #4968
• documentation improvements Error message when (ab)using frollmean on list with adaptive = TRUE, and notes on documentation  #5306

Closes #5438, #4968 and #5306.

Benchmarks

• adaptive left aligned rolling max (algo="exact" only); benchmark code in NEWS.md file

#Unit: milliseconds
#       expr        min         lq       mean     median         uq        max neval
#   baser(x) 3094.88357 3097.84966 3186.74832 3163.58050 3251.66753 3370.33785    10
#      sj(x) 2221.55456 2255.12083 2306.61382 2303.47883 2346.70293 2412.62975    10
#   frmax(x)   17.45124   24.16809   28.10062   28.58153   32.79802   34.83941    10
# frapply(x)  272.07830  316.47060  366.94771  396.23566  416.06699  421.38701    10

• simple rolling max use case against very fast roll library

library(roll)
library(data.table)
set.seed(108)
x = rnorm(1e8)
n = 1e5
system.time(
  a1 <- roll_max(x, n)
)
#   user  system elapsed 
#  1.144   0.266   1.411
system.time(
  a2 <- frollmax(x, n)
)
#   user  system elapsed 
#  0.178   0.264   0.443
all.equal(a1, a2)
#[1] TRUE

This work was supported by the Medical Research Council, UK [grant number 1573].

[tdhock]

lgtm

[jangorecki]

For now I have reverted atime test script and merged fresh master, to have this PR in a ready-to-merge state.
I am of course open to add atime script but I am not sure if it's feasible now, as per in-line comments in previous commits.

[tdhock]

that is fine, no need to add atime tests, lgtm.

[ben-schwen]

Some small things but overall LGTM