Using row index `.I` for joins as an argument to `on=`

Start with the reproducible R example in the issue and review issue #1494, which the report identifies as related. Done means joins can use the row index through on= without creating a temporary row_index column, either by fixing .I behavior or introducing a clearer .ROWI/.ROWINDEX alternative.

library(data.table)
dt1 = CJ(a=1:5, b=c("yes","no"))
# Key: <a, b>
# Index: <b>
#         a      b
#     <int> <char>
#  1:     1     no
#  2:     1    yes
#  3:     2     no
#  4:     2    yes
#  5:     3     no
#  6:     3    yes
#  7:     4     no
#  8:     4    yes
#  9:     5     no
# 10:     5    yes
dt2 = dt1[b == "yes", list(.RESULT = list(seq(0, a))), by=.I][,unlist(.RESULT), by=I]
#         I    V1
#     <int> <int>
#  1:     2     0
#  2:     2     1
#  3:     4     0
#  4:     4     1
#  5:     4     2
#  6:     6     0
#  7:     6     1
#  8:     6     2
#  9:     6     3
# 10:     8     0
# 11:     8     1
# 12:     8     2
# 13:     8     3
# 14:     8     4
# 15:    10     0
# 16:    10     1
# 17:    10     2
# 18:    10     3
# 19:    10     4
# 20:    10     5
#         I    V1
## This is what I want to do but it gives an error
# dt1 <- dt1[dt2, on=list(.I="I")]
# Error in colnamesInt(x, names(on), check_dups = FALSE) : 
#  argument specifying columns received non-existing column(s): cols[1]='.I'
## This is what I need to do but it is so much longer and complicated (and probably slower)
tmp <- copy(dt1)
tmp <- tmp[, row_index:=.I][dt2, on = list(row_index=I)]
tmp[, row_index:=NULL]
dt1 <- tmp
rm(tmp)
dt1
# Key: <a, b>
#         a      b    V1
#     <int> <char> <int>
#  1:     1    yes     0
#  2:     1    yes     1
#  3:     2    yes     0
#  4:     2    yes     1
#  5:     2    yes     2
#  6:     3    yes     0
#  7:     3    yes     1
#  8:     3    yes     2
#  9:     3    yes     3
# 10:     4    yes     0
# 11:     4    yes     1
# 12:     4    yes     2
# 13:     4    yes     3
# 14:     4    yes     4
# 15:     5    yes     0
# 16:     5    yes     1
# 17:     5    yes     2
# 18:     5    yes     3
# 19:     5    yes     4
# 20:     5    yes     5
#         a      b    V1

To make this work, issue #1494 probably needs to be fixed or introduce a different .ROWI that doesn't have .I's behavior.

(.ROWI or .ROWINDEX is anyway clearer than .I)