Replace equality definition on ObserverExpression

[mdickinson]

The current ObserverExpression implements equality via _as_graphs, I suspect purely as a convenience for testing purposes. This PR replaces that equality check with a simple structural equality. It also:

• makes all expressions hashable, to make caching easier
• adds convenience __repr__ implementations
• renames the observer attribute to _observer, for consistency with the other expression classes, to emphasize that it's not used outside this module, and to discourage users from modifying the value
• fixes one (recently-introduced) test

[mdickinson]

Actually, I think we can do a little better on the __repr__. Let me remove it from this PR, and re-introduce in a separate one.

[corranwebster]

Looks good.

Some small concerns about performance if the equality test becomes critical path since it is effectively recursive, but if it becomes a problem we can fix, and it's probably no worse than testing graph equality.

[mdickinson]

Since this is partly about performance, I should probably include some code timings. Here's a (successful) equality test on a reasonably complicated but still somewhat realistic expression:

On main:

(traits) mdickinson@mirzakhani traits % ipython
Python 3.9.7 (default, Sep  1 2021, 04:00:27) 
Type 'copyright', 'credits' or 'license' for more information
IPython 7.27.0 -- An enhanced Interactive Python. Type '?' for help.

In [1]: from traits.observation.api import *

In [2]: expr1 = parse("a,[b,c].items.updated,d:e,f")

In [3]: expr2 = parse("a,[b,c].items.updated,d:e,f")

In [4]: expr1 == expr2
Out[4]: True

In [5]: %timeit expr1 == expr2
100 µs ± 2.7 µs per loop (mean ± std. dev. of 7 runs, 10000 loops each)

and on this branch, with the same exact definitions of expr1 and expr2:

In [7]: %timeit expr1 == expr2
8.54 µs ± 171 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each)

Failed equality tests should be much faster. Again, on main:

In [6]: expr1 = parse("a,[b,c].items.updated,d:e,f")

In [7]: expr2 = parse("[a,b,c].items.updated,d:e,f")

In [8]: expr1 == expr2
Out[8]: False

In [9]: %timeit expr1 == expr2
62.7 µs ± 1.29 µs per loop (mean ± std. dev. of 7 runs, 10000 loops each)

and the same test on this branch:

(traits) mdickinson@mirzakhani traits % ipython
Python 3.9.7 (default, Sep  1 2021, 04:00:27) 
Type 'copyright', 'credits' or 'license' for more information
IPython 7.27.0 -- An enhanced Interactive Python. Type '?' for help.

In [1]: from traits.observation.api import parse

In [2]: expr1 = parse("a,[b,c].items.updated,d:e,f")

In [3]: expr2 = parse("[a,b,c].items.updated,d:e,f")

In [4]: %timeit expr1 == expr2
652 ns ± 14 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)