Prefer use of peek_token_ref over peek_token where valid

[alexander-beedie]

Started looking at performance optimisation opportunities that don't require structural or API changes - have identified several, and will make a series of PRs that each cover one change, to keep things clean/focused and easy to review.

Hopefully there will be a nice cumulative impact 👍

Description

Where valid, the following updates have been made to avoid unnecessary clones, by preferring the "ref" variant of several core parsing functions:

• peek_token → peek_token_ref (~200 updates)
• expected → expected_ref (~100 updates)

Results

Local testing with sqlparser_bench tests showed the following improvements:

┌──────────────────────────────┬───────────┬───────────┬────────┐
│ benchmark                    │ main      │ this pr   │ change │
├──────────────────────────────┼───────────┼───────────┼────────┤
│ sqlparser::select            │ 2.704 µs  │ 2.667 µs  │ -0.9%  │
├──────────────────────────────┼───────────┼───────────┼────────┤
│ sqlparser::with_select       │ 12.893 µs │ 12.642 µs │ -1.7%  │
├──────────────────────────────┼───────────┼───────────┼────────┤
│ parse_large_statement        │ 4.449 ms  │ 4.310 ms  │ -3.1%  │
├──────────────────────────────┼───────────┼───────────┼────────┤
│ select_100_columns           │ 68.14 µs  │ 66.42 µs  │ -3.4%  │
├──────────────────────────────┼───────────┼───────────┼────────┤
│ select_100_qualified_columns │ 141.35 µs │ 138.80 µs │ -1.6%  │
└──────────────────────────────┴───────────┴───────────┴────────┘

[iffyio]

LGTM! Thanks @alexander-beedie!