Optimize ID/tokens code

[MetRonnie]

Profiling revealed Tokens computed properties and detokenise() were being called many times and so was acting as a bit of a bottleneck (within the bottleneck of data store manager's increment_graph_window() - #5315).

• Remove a couple of unnecessary nested comprehensions
• Precompute certain sets

Partially addresses #5315 & #5183

Context

Click to see `flow.cylc`

#!Jinja2

{% set TASKS = TASKS | default(100) %}
{% set CYCLES = CYCLES | default(1) %}
{% set SLEEP = SLEEP | default(1) %}

[meta]
    description = """
        Example scaling workflow with {{ (TASKS * 2) + 2 }} tasks and {{ (TASKS ** 2) + (TASKS * 2) }}
        dependencies per cycle which runs for {{ CYCLES }} cycles.
    """

[task parameters]
    x = 1..{{ TASKS }}
    y = 1..{{ TASKS }}

[scheduling]
    cycling mode = integer
    initial cycle point = 1
    final cycle point = {{ CYCLES }}
    [[graph]]
        P1 = """
            d[-P1] => a => b<x> => c<y> => d
        """

[runtime]
    [[b<x>, c<y>]]
        script = sleep {{ SLEEP }}
    [[a, d]]

$ cylc play efficiency/ --profile -s 'TASKS=16'

Expanded to edge_id:

After

Check list

• I have read CONTRIBUTING.md and added my name as a Code Contributor.
• Contains logically grouped changes (else tidy your branch by rebase).
• Does not contain off-topic changes (use other PRs for other changes).
• No dependency changes
• No new tests needed
• No changelog entry - presumably will be part of a bunch of efficiency improvements for 8.1.1
• No docs
• If this is a bug fix, PR should be raised against the relevant ?.?.x branch.

[oliver-sanders]

LGTM, 16% passive speedup for the given example!

[hjoliver]

LGTM, nice speedup 👍