This PR improves the token completion with the following changes:

1. Since Add '(' to completion trigger characters #369 allows ( to trigger completion to in accompany with Add initial pipeline completion support vscode-R#530, the current token_completion may produce too many symbols once ( triggers completion. This PR makes token_completion only show symbols after $ and SYMBOL_SUB symbols like foo in mutate(foo = bar) when trigger token is empty (e.g. foo$ and fun(), which are the most frequent ways to introduce new data variables.
2. If token is not empty, then all symbols from the document that start with token are searched, which is the original method.

This makes it possible to work with the following examples:

library(data.table)
library(dplyr)
library(tibble)

tbl1 <- tibble(
  Name = c("Product1", "Product2", "Product3"),
  Quality = c("Good", "Good", "Best"),
  Score = c(8, 9, 10),
  Type = c("B", "B", "A")
)

tbl1 %>%
  mutate(Score_mean = mean(Score)) %>%
  select(|

x <- iris
x %>%
  mutate(score1 = Petal.Length + Petal.Width, score2 = Petal.Length * Petal.Width) %>%
  filter(|


df <- list(
  test1 = 1,
  test2 = 2
)
df$test3 <- 3
df$te|


dt <- data.table(var1 = 1:10, var2 = rnorm(10))
dt[, var3 := var1 + var2]
dt[, var4 := var|]

where | is the cursor and the completion should work as expected to complete the tokens from the document.