Second README pass: cut explanation the reader does not act on

[lucapinello]

Prompted by a fair callout: "A volume sold as 100 GB is 93.1 GiB, so it fits with ~2 GiB spare" is a units
lesson, not something anyone provisioning a disk uses. The first pass moved caveats out of the reading path;
this one removes over-explanation in it.

One rule throughout: keep what changes what a user does, cut what only explains how we know it or why we
chose it.

Lectures removed

cut	why
decimal-vs-binary lesson on the free-disk bullet	you need the number, not the arithmetic
~15 lines on why conda envs are large — pip hardlinking, du -sc soname counting	reduced to the two facts you act on: macOS/CPU installs are smaller, and --oracle + the --no-* flags skip what you don't need
a note explaining that two of our own tables size the same files in different units	fixing our inconsistency is not the reader's job
what an earlier audit measured on other hardware, in the equivalence claim	mine, from yesterday; current numbers are what matter
the history of the downloader's previous "only fetch when missing" rule	the rule is useful, its predecessor is not
my postmortem in Caveats about why our test missed the splice bug	you need the state and the action, not our test-design retrospective

Duplications removed — structurally the same problem

• A BedGraph "example" that was three comments and a pointer to the real recipe.
• Two ways to load the identical genome (get_genome, then GenomeManager doing the same thing).
• Core concepts re-defining "Oracle" and "Track" — both already in Key terms.
• Key terms re-defining "Conversational genomics" — from the paragraph three lines above it.
• The credentials/conda-env rule stated in two separate places.

Plus two structural fixes: an aside was splitting the two-item token list, and "the wrong place for 85 GB"
was a third stale copy of a total that had moved to 87.

Cumulative since v0.7.5

	at v0.7.5	now
aside text	12,229 chars	6,409 (−48%)
longest aside	1,692	704
prose lines > 420 chars	17	8
README	99,211 chars	91,959

One guard deleted rather than satisfied

test_the_tldr_install_size_agrees_with_the_disk_table — I wrote it yesterday to keep the TLDR's install
size in sync with the disk table. The TLDR no longer states an install size, so there is nothing to keep in
sync. Removing the duplicate removed the reason for the guard; re-adding the number to keep a test green
would have been backwards.

Fast suite 2,144 passed / 35 skipped. The skip count rose by one because the changelog-vs-tag guard
correctly skips on an untagged branch HEAD — not lost coverage, checked rather than assumed.